2 * Copyright (C) 2011 - David Goulet <david.goulet@polymtl.ca>
3 * Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
4 * 2013 - Jérémie Galarneau <jeremie.galarneau@efficios.com>
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License, version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 * GNU General Public License for more details.
15 * You should have received a copy of the GNU General Public License along
16 * with this program; if not, write to the Free Software Foundation, Inc.,
17 * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
32 #include <sys/mount.h>
33 #include <sys/resource.h>
34 #include <sys/socket.h>
36 #include <sys/types.h>
38 #include <urcu/uatomic.h>
42 #include <common/common.h>
43 #include <common/compat/socket.h>
44 #include <common/compat/getenv.h>
45 #include <common/defaults.h>
46 #include <common/kernel-consumer/kernel-consumer.h>
47 #include <common/futex.h>
48 #include <common/relayd/relayd.h>
49 #include <common/utils.h>
50 #include <common/daemonize.h>
51 #include <common/config/session-config.h>
53 #include "lttng-sessiond.h"
54 #include "buffer-registry.h"
61 #include "kernel-consumer.h"
65 #include "ust-consumer.h"
68 #include "health-sessiond.h"
69 #include "testpoint.h"
70 #include "ust-thread.h"
71 #include "agent-thread.h"
73 #include "load-session-thread.h"
74 #include "notification-thread.h"
75 #include "notification-thread-commands.h"
78 #include "ht-cleanup.h"
79 #include "sessiond-config.h"
81 static const char *help_msg
=
82 #ifdef LTTNG_EMBED_HELP
83 #include <lttng-sessiond.8.h>
90 static pid_t ppid
; /* Parent PID for --sig-parent option */
91 static pid_t child_ppid
; /* Internal parent PID use with daemonize. */
92 static int lockfile_fd
= -1;
94 /* Set to 1 when a SIGUSR1 signal is received. */
95 static int recv_child_signal
;
98 * Consumer daemon specific control data. Every value not initialized here is
99 * set to 0 by the static definition.
101 static struct consumer_data kconsumer_data
= {
102 .type
= LTTNG_CONSUMER_KERNEL
,
105 .channel_monitor_pipe
= -1,
106 .pid_mutex
= PTHREAD_MUTEX_INITIALIZER
,
107 .lock
= PTHREAD_MUTEX_INITIALIZER
,
108 .cond
= PTHREAD_COND_INITIALIZER
,
109 .cond_mutex
= PTHREAD_MUTEX_INITIALIZER
,
111 static struct consumer_data ustconsumer64_data
= {
112 .type
= LTTNG_CONSUMER64_UST
,
115 .channel_monitor_pipe
= -1,
116 .pid_mutex
= PTHREAD_MUTEX_INITIALIZER
,
117 .lock
= PTHREAD_MUTEX_INITIALIZER
,
118 .cond
= PTHREAD_COND_INITIALIZER
,
119 .cond_mutex
= PTHREAD_MUTEX_INITIALIZER
,
121 static struct consumer_data ustconsumer32_data
= {
122 .type
= LTTNG_CONSUMER32_UST
,
125 .channel_monitor_pipe
= -1,
126 .pid_mutex
= PTHREAD_MUTEX_INITIALIZER
,
127 .lock
= PTHREAD_MUTEX_INITIALIZER
,
128 .cond
= PTHREAD_COND_INITIALIZER
,
129 .cond_mutex
= PTHREAD_MUTEX_INITIALIZER
,
132 /* Command line options */
133 static const struct option long_options
[] = {
134 { "client-sock", required_argument
, 0, 'c' },
135 { "apps-sock", required_argument
, 0, 'a' },
136 { "kconsumerd-cmd-sock", required_argument
, 0, '\0' },
137 { "kconsumerd-err-sock", required_argument
, 0, '\0' },
138 { "ustconsumerd32-cmd-sock", required_argument
, 0, '\0' },
139 { "ustconsumerd32-err-sock", required_argument
, 0, '\0' },
140 { "ustconsumerd64-cmd-sock", required_argument
, 0, '\0' },
141 { "ustconsumerd64-err-sock", required_argument
, 0, '\0' },
142 { "consumerd32-path", required_argument
, 0, '\0' },
143 { "consumerd32-libdir", required_argument
, 0, '\0' },
144 { "consumerd64-path", required_argument
, 0, '\0' },
145 { "consumerd64-libdir", required_argument
, 0, '\0' },
146 { "daemonize", no_argument
, 0, 'd' },
147 { "background", no_argument
, 0, 'b' },
148 { "sig-parent", no_argument
, 0, 'S' },
149 { "help", no_argument
, 0, 'h' },
150 { "group", required_argument
, 0, 'g' },
151 { "version", no_argument
, 0, 'V' },
152 { "quiet", no_argument
, 0, 'q' },
153 { "verbose", no_argument
, 0, 'v' },
154 { "verbose-consumer", no_argument
, 0, '\0' },
155 { "no-kernel", no_argument
, 0, '\0' },
156 { "pidfile", required_argument
, 0, 'p' },
157 { "agent-tcp-port", required_argument
, 0, '\0' },
158 { "config", required_argument
, 0, 'f' },
159 { "load", required_argument
, 0, 'l' },
160 { "kmod-probes", required_argument
, 0, '\0' },
161 { "extra-kmod-probes", required_argument
, 0, '\0' },
165 struct sessiond_config config
;
167 /* Command line options to ignore from configuration file */
168 static const char *config_ignore_options
[] = { "help", "version", "config" };
170 /* Shared between threads */
171 static int dispatch_thread_exit
;
173 /* Sockets and FDs */
174 static int client_sock
= -1;
175 static int apps_sock
= -1;
176 int kernel_tracer_fd
= -1;
177 static int kernel_poll_pipe
[2] = { -1, -1 };
180 * Quit pipe for all threads. This permits a single cancellation point
181 * for all threads when receiving an event on the pipe.
183 static int thread_quit_pipe
[2] = { -1, -1 };
186 * This pipe is used to inform the thread managing application communication
187 * that a command is queued and ready to be processed.
189 static int apps_cmd_pipe
[2] = { -1, -1 };
191 int apps_cmd_notify_pipe
[2] = { -1, -1 };
193 /* Pthread, Mutexes and Semaphores */
194 static pthread_t apps_thread
;
195 static pthread_t apps_notify_thread
;
196 static pthread_t reg_apps_thread
;
197 static pthread_t client_thread
;
198 static pthread_t kernel_thread
;
199 static pthread_t dispatch_thread
;
200 static pthread_t health_thread
;
201 static pthread_t ht_cleanup_thread
;
202 static pthread_t agent_reg_thread
;
203 static pthread_t load_session_thread
;
204 static pthread_t notification_thread
;
207 * UST registration command queue. This queue is tied with a futex and uses a N
208 * wakers / 1 waiter implemented and detailed in futex.c/.h
210 * The thread_registration_apps and thread_dispatch_ust_registration uses this
211 * queue along with the wait/wake scheme. The thread_manage_apps receives down
212 * the line new application socket and monitors it for any I/O error or clean
213 * close that triggers an unregistration of the application.
215 static struct ust_cmd_queue ust_cmd_queue
;
218 * Pointer initialized before thread creation.
220 * This points to the tracing session list containing the session count and a
221 * mutex lock. The lock MUST be taken if you iterate over the list. The lock
222 * MUST NOT be taken if you call a public function in session.c.
224 * The lock is nested inside the structure: session_list_ptr->lock. Please use
225 * session_lock_list and session_unlock_list for lock acquisition.
227 static struct ltt_session_list
*session_list_ptr
;
229 int ust_consumerd64_fd
= -1;
230 int ust_consumerd32_fd
= -1;
232 static const char *module_proc_lttng
= "/proc/lttng";
235 * Consumer daemon state which is changed when spawning it, killing it or in
236 * case of a fatal error.
238 enum consumerd_state
{
239 CONSUMER_STARTED
= 1,
240 CONSUMER_STOPPED
= 2,
245 * This consumer daemon state is used to validate if a client command will be
246 * able to reach the consumer. If not, the client is informed. For instance,
247 * doing a "lttng start" when the consumer state is set to ERROR will return an
248 * error to the client.
250 * The following example shows a possible race condition of this scheme:
252 * consumer thread error happens
254 * client cmd checks state -> still OK
255 * consumer thread exit, sets error
256 * client cmd try to talk to consumer
259 * However, since the consumer is a different daemon, we have no way of making
260 * sure the command will reach it safely even with this state flag. This is why
261 * we consider that up to the state validation during command processing, the
262 * command is safe. After that, we can not guarantee the correctness of the
263 * client request vis-a-vis the consumer.
265 static enum consumerd_state ust_consumerd_state
;
266 static enum consumerd_state kernel_consumerd_state
;
268 /* Set in main() with the current page size. */
271 /* Application health monitoring */
272 struct health_app
*health_sessiond
;
274 /* Am I root or not. */
275 int is_root
; /* Set to 1 if the daemon is running as root */
277 const char * const config_section_name
= "sessiond";
279 /* Load session thread information to operate. */
280 struct load_session_thread_data
*load_info
;
282 /* Notification thread handle. */
283 struct notification_thread_handle
*notification_thread_handle
;
285 /* Global hash tables */
286 struct lttng_ht
*agent_apps_ht_by_sock
= NULL
;
289 * Whether sessiond is ready for commands/notification channel/health check
291 * NR_LTTNG_SESSIOND_READY must match the number of calls to
292 * sessiond_notify_ready().
294 #define NR_LTTNG_SESSIOND_READY 4
295 int lttng_sessiond_ready
= NR_LTTNG_SESSIOND_READY
;
297 int sessiond_check_thread_quit_pipe(int fd
, uint32_t events
)
299 return (fd
== thread_quit_pipe
[0] && (events
& LPOLLIN
)) ? 1 : 0;
302 /* Notify parents that we are ready for cmd and health check */
304 void sessiond_notify_ready(void)
306 if (uatomic_sub_return(<tng_sessiond_ready
, 1) == 0) {
308 * Notify parent pid that we are ready to accept command
309 * for client side. This ppid is the one from the
310 * external process that spawned us.
312 if (config
.sig_parent
) {
317 * Notify the parent of the fork() process that we are
320 if (config
.daemonize
|| config
.background
) {
321 kill(child_ppid
, SIGUSR1
);
327 int __sessiond_set_thread_pollset(struct lttng_poll_event
*events
, size_t size
,
334 ret
= lttng_poll_create(events
, size
, LTTNG_CLOEXEC
);
340 ret
= lttng_poll_add(events
, a_pipe
[0], LPOLLIN
| LPOLLERR
);
352 * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
354 int sessiond_set_thread_pollset(struct lttng_poll_event
*events
, size_t size
)
356 return __sessiond_set_thread_pollset(events
, size
, thread_quit_pipe
);
360 * Init thread quit pipe.
362 * Return -1 on error or 0 if all pipes are created.
364 static int __init_thread_quit_pipe(int *a_pipe
)
370 PERROR("thread quit pipe");
374 for (i
= 0; i
< 2; i
++) {
375 ret
= fcntl(a_pipe
[i
], F_SETFD
, FD_CLOEXEC
);
386 static int init_thread_quit_pipe(void)
388 return __init_thread_quit_pipe(thread_quit_pipe
);
392 * Stop all threads by closing the thread quit pipe.
394 static void stop_threads(void)
398 /* Stopping all threads */
399 DBG("Terminating all threads");
400 ret
= notify_thread_pipe(thread_quit_pipe
[1]);
402 ERR("write error on thread quit pipe");
405 /* Dispatch thread */
406 CMM_STORE_SHARED(dispatch_thread_exit
, 1);
407 futex_nto1_wake(&ust_cmd_queue
.futex
);
411 * Close every consumer sockets.
413 static void close_consumer_sockets(void)
417 if (kconsumer_data
.err_sock
>= 0) {
418 ret
= close(kconsumer_data
.err_sock
);
420 PERROR("kernel consumer err_sock close");
423 if (ustconsumer32_data
.err_sock
>= 0) {
424 ret
= close(ustconsumer32_data
.err_sock
);
426 PERROR("UST consumerd32 err_sock close");
429 if (ustconsumer64_data
.err_sock
>= 0) {
430 ret
= close(ustconsumer64_data
.err_sock
);
432 PERROR("UST consumerd64 err_sock close");
435 if (kconsumer_data
.cmd_sock
>= 0) {
436 ret
= close(kconsumer_data
.cmd_sock
);
438 PERROR("kernel consumer cmd_sock close");
441 if (ustconsumer32_data
.cmd_sock
>= 0) {
442 ret
= close(ustconsumer32_data
.cmd_sock
);
444 PERROR("UST consumerd32 cmd_sock close");
447 if (ustconsumer64_data
.cmd_sock
>= 0) {
448 ret
= close(ustconsumer64_data
.cmd_sock
);
450 PERROR("UST consumerd64 cmd_sock close");
453 if (kconsumer_data
.channel_monitor_pipe
>= 0) {
454 ret
= close(kconsumer_data
.channel_monitor_pipe
);
456 PERROR("kernel consumer channel monitor pipe close");
459 if (ustconsumer32_data
.channel_monitor_pipe
>= 0) {
460 ret
= close(ustconsumer32_data
.channel_monitor_pipe
);
462 PERROR("UST consumerd32 channel monitor pipe close");
465 if (ustconsumer64_data
.channel_monitor_pipe
>= 0) {
466 ret
= close(ustconsumer64_data
.channel_monitor_pipe
);
468 PERROR("UST consumerd64 channel monitor pipe close");
474 * Wait on consumer process termination.
476 * Need to be called with the consumer data lock held or from a context
477 * ensuring no concurrent access to data (e.g: cleanup).
479 static void wait_consumer(struct consumer_data
*consumer_data
)
484 if (consumer_data
->pid
<= 0) {
488 DBG("Waiting for complete teardown of consumerd (PID: %d)",
490 ret
= waitpid(consumer_data
->pid
, &status
, 0);
492 PERROR("consumerd waitpid pid: %d", consumer_data
->pid
)
493 } else if (!WIFEXITED(status
)) {
494 ERR("consumerd termination with error: %d",
497 consumer_data
->pid
= 0;
501 * Cleanup the session daemon's data structures.
503 static void sessiond_cleanup(void)
506 struct ltt_session
*sess
, *stmp
;
508 DBG("Cleanup sessiond");
511 * Close the thread quit pipe. It has already done its job,
512 * since we are now called.
514 utils_close_pipe(thread_quit_pipe
);
517 * If config.pid_file_path.value is undefined, the default file will be
518 * wiped when removing the rundir.
520 if (config
.pid_file_path
.value
) {
521 ret
= remove(config
.pid_file_path
.value
);
523 PERROR("remove pidfile %s", config
.pid_file_path
.value
);
527 DBG("Removing sessiond and consumerd content of directory %s",
528 config
.rundir
.value
);
531 DBG("Removing %s", config
.pid_file_path
.value
);
532 (void) unlink(config
.pid_file_path
.value
);
534 DBG("Removing %s", config
.agent_port_file_path
.value
);
535 (void) unlink(config
.agent_port_file_path
.value
);
538 DBG("Removing %s", kconsumer_data
.err_unix_sock_path
);
539 (void) unlink(kconsumer_data
.err_unix_sock_path
);
541 DBG("Removing directory %s", config
.kconsumerd_path
.value
);
542 (void) rmdir(config
.kconsumerd_path
.value
);
544 /* ust consumerd 32 */
545 DBG("Removing %s", config
.consumerd32_err_unix_sock_path
.value
);
546 (void) unlink(config
.consumerd32_err_unix_sock_path
.value
);
548 DBG("Removing directory %s", config
.consumerd32_path
.value
);
549 (void) rmdir(config
.consumerd32_path
.value
);
551 /* ust consumerd 64 */
552 DBG("Removing %s", config
.consumerd64_err_unix_sock_path
.value
);
553 (void) unlink(config
.consumerd64_err_unix_sock_path
.value
);
555 DBG("Removing directory %s", config
.consumerd64_path
.value
);
556 (void) rmdir(config
.consumerd64_path
.value
);
558 DBG("Cleaning up all sessions");
560 /* Destroy session list mutex */
561 if (session_list_ptr
!= NULL
) {
562 pthread_mutex_destroy(&session_list_ptr
->lock
);
564 /* Cleanup ALL session */
565 cds_list_for_each_entry_safe(sess
, stmp
,
566 &session_list_ptr
->head
, list
) {
567 cmd_destroy_session(sess
, kernel_poll_pipe
[1]);
571 wait_consumer(&kconsumer_data
);
572 wait_consumer(&ustconsumer64_data
);
573 wait_consumer(&ustconsumer32_data
);
575 DBG("Cleaning up all agent apps");
576 agent_app_ht_clean();
578 DBG("Closing all UST sockets");
579 ust_app_clean_list();
580 buffer_reg_destroy_registries();
582 if (is_root
&& !config
.no_kernel
) {
583 DBG2("Closing kernel fd");
584 if (kernel_tracer_fd
>= 0) {
585 ret
= close(kernel_tracer_fd
);
590 DBG("Unloading kernel modules");
591 modprobe_remove_lttng_all();
595 close_consumer_sockets();
598 load_session_destroy_data(load_info
);
603 * Cleanup lock file by deleting it and finaly closing it which will
604 * release the file system lock.
606 if (lockfile_fd
>= 0) {
607 ret
= remove(config
.lock_file_path
.value
);
609 PERROR("remove lock file");
611 ret
= close(lockfile_fd
);
613 PERROR("close lock file");
618 * We do NOT rmdir rundir because there are other processes
619 * using it, for instance lttng-relayd, which can start in
620 * parallel with this teardown.
625 * Cleanup the daemon's option data structures.
627 static void sessiond_cleanup_options(void)
629 DBG("Cleaning up options");
631 sessiond_config_fini(&config
);
633 run_as_destroy_worker();
637 * Send data on a unix socket using the liblttsessiondcomm API.
639 * Return lttcomm error code.
641 static int send_unix_sock(int sock
, void *buf
, size_t len
)
643 /* Check valid length */
648 return lttcomm_send_unix_sock(sock
, buf
, len
);
652 * Free memory of a command context structure.
654 static void clean_command_ctx(struct command_ctx
**cmd_ctx
)
656 DBG("Clean command context structure");
658 if ((*cmd_ctx
)->llm
) {
659 free((*cmd_ctx
)->llm
);
661 if ((*cmd_ctx
)->lsm
) {
662 free((*cmd_ctx
)->lsm
);
670 * Notify UST applications using the shm mmap futex.
672 static int notify_ust_apps(int active
)
676 DBG("Notifying applications of session daemon state: %d", active
);
678 /* See shm.c for this call implying mmap, shm and futex calls */
679 wait_shm_mmap
= shm_ust_get_mmap(config
.wait_shm_path
.value
, is_root
);
680 if (wait_shm_mmap
== NULL
) {
684 /* Wake waiting process */
685 futex_wait_update((int32_t *) wait_shm_mmap
, active
);
687 /* Apps notified successfully */
695 * Setup the outgoing data buffer for the response (llm) by allocating the
696 * right amount of memory and copying the original information from the lsm
699 * Return 0 on success, negative value on error.
701 static int setup_lttng_msg(struct command_ctx
*cmd_ctx
,
702 const void *payload_buf
, size_t payload_len
,
703 const void *cmd_header_buf
, size_t cmd_header_len
)
706 const size_t header_len
= sizeof(struct lttcomm_lttng_msg
);
707 const size_t cmd_header_offset
= header_len
;
708 const size_t payload_offset
= cmd_header_offset
+ cmd_header_len
;
709 const size_t total_msg_size
= header_len
+ cmd_header_len
+ payload_len
;
711 cmd_ctx
->llm
= zmalloc(total_msg_size
);
713 if (cmd_ctx
->llm
== NULL
) {
719 /* Copy common data */
720 cmd_ctx
->llm
->cmd_type
= cmd_ctx
->lsm
->cmd_type
;
721 cmd_ctx
->llm
->pid
= cmd_ctx
->lsm
->domain
.attr
.pid
;
722 cmd_ctx
->llm
->cmd_header_size
= cmd_header_len
;
723 cmd_ctx
->llm
->data_size
= payload_len
;
724 cmd_ctx
->lttng_msg_size
= total_msg_size
;
726 /* Copy command header */
727 if (cmd_header_len
) {
728 memcpy(((uint8_t *) cmd_ctx
->llm
) + cmd_header_offset
, cmd_header_buf
,
734 memcpy(((uint8_t *) cmd_ctx
->llm
) + payload_offset
, payload_buf
,
743 * Version of setup_lttng_msg() without command header.
745 static int setup_lttng_msg_no_cmd_header(struct command_ctx
*cmd_ctx
,
746 void *payload_buf
, size_t payload_len
)
748 return setup_lttng_msg(cmd_ctx
, payload_buf
, payload_len
, NULL
, 0);
751 * Update the kernel poll set of all channel fd available over all tracing
752 * session. Add the wakeup pipe at the end of the set.
754 static int update_kernel_poll(struct lttng_poll_event
*events
)
757 struct ltt_session
*session
;
758 struct ltt_kernel_channel
*channel
;
760 DBG("Updating kernel poll set");
763 cds_list_for_each_entry(session
, &session_list_ptr
->head
, list
) {
764 session_lock(session
);
765 if (session
->kernel_session
== NULL
) {
766 session_unlock(session
);
770 cds_list_for_each_entry(channel
,
771 &session
->kernel_session
->channel_list
.head
, list
) {
772 /* Add channel fd to the kernel poll set */
773 ret
= lttng_poll_add(events
, channel
->fd
, LPOLLIN
| LPOLLRDNORM
);
775 session_unlock(session
);
778 DBG("Channel fd %d added to kernel set", channel
->fd
);
780 session_unlock(session
);
782 session_unlock_list();
787 session_unlock_list();
792 * Find the channel fd from 'fd' over all tracing session. When found, check
793 * for new channel stream and send those stream fds to the kernel consumer.
795 * Useful for CPU hotplug feature.
797 static int update_kernel_stream(struct consumer_data
*consumer_data
, int fd
)
800 struct ltt_session
*session
;
801 struct ltt_kernel_session
*ksess
;
802 struct ltt_kernel_channel
*channel
;
804 DBG("Updating kernel streams for channel fd %d", fd
);
807 cds_list_for_each_entry(session
, &session_list_ptr
->head
, list
) {
808 session_lock(session
);
809 if (session
->kernel_session
== NULL
) {
810 session_unlock(session
);
813 ksess
= session
->kernel_session
;
815 cds_list_for_each_entry(channel
,
816 &ksess
->channel_list
.head
, list
) {
817 struct lttng_ht_iter iter
;
818 struct consumer_socket
*socket
;
820 if (channel
->fd
!= fd
) {
823 DBG("Channel found, updating kernel streams");
824 ret
= kernel_open_channel_stream(channel
);
828 /* Update the stream global counter */
829 ksess
->stream_count_global
+= ret
;
832 * Have we already sent fds to the consumer? If yes, it
833 * means that tracing is started so it is safe to send
834 * our updated stream fds.
836 if (ksess
->consumer_fds_sent
!= 1
837 || ksess
->consumer
== NULL
) {
843 cds_lfht_for_each_entry(ksess
->consumer
->socks
->ht
,
844 &iter
.iter
, socket
, node
.node
) {
845 pthread_mutex_lock(socket
->lock
);
846 ret
= kernel_consumer_send_channel_stream(socket
,
848 session
->output_traces
? 1 : 0);
849 pthread_mutex_unlock(socket
->lock
);
857 session_unlock(session
);
859 session_unlock_list();
863 session_unlock(session
);
864 session_unlock_list();
869 * For each tracing session, update newly registered apps. The session list
870 * lock MUST be acquired before calling this.
872 static void update_ust_app(int app_sock
)
874 struct ltt_session
*sess
, *stmp
;
876 /* Consumer is in an ERROR state. Stop any application update. */
877 if (uatomic_read(&ust_consumerd_state
) == CONSUMER_ERROR
) {
878 /* Stop the update process since the consumer is dead. */
882 /* For all tracing session(s) */
883 cds_list_for_each_entry_safe(sess
, stmp
, &session_list_ptr
->head
, list
) {
887 if (!sess
->ust_session
) {
892 assert(app_sock
>= 0);
893 app
= ust_app_find_by_sock(app_sock
);
896 * Application can be unregistered before so
897 * this is possible hence simply stopping the
900 DBG3("UST app update failed to find app sock %d",
904 ust_app_global_update(sess
->ust_session
, app
);
908 session_unlock(sess
);
913 * This thread manage event coming from the kernel.
915 * Features supported in this thread:
918 static void *thread_manage_kernel(void *data
)
920 int ret
, i
, pollfd
, update_poll_flag
= 1, err
= -1;
921 uint32_t revents
, nb_fd
;
923 struct lttng_poll_event events
;
925 DBG("[thread] Thread manage kernel started");
927 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_KERNEL
);
930 * This first step of the while is to clean this structure which could free
931 * non NULL pointers so initialize it before the loop.
933 lttng_poll_init(&events
);
935 if (testpoint(sessiond_thread_manage_kernel
)) {
936 goto error_testpoint
;
939 health_code_update();
941 if (testpoint(sessiond_thread_manage_kernel_before_loop
)) {
942 goto error_testpoint
;
946 health_code_update();
948 if (update_poll_flag
== 1) {
949 /* Clean events object. We are about to populate it again. */
950 lttng_poll_clean(&events
);
952 ret
= sessiond_set_thread_pollset(&events
, 2);
954 goto error_poll_create
;
957 ret
= lttng_poll_add(&events
, kernel_poll_pipe
[0], LPOLLIN
);
962 /* This will add the available kernel channel if any. */
963 ret
= update_kernel_poll(&events
);
967 update_poll_flag
= 0;
970 DBG("Thread kernel polling");
972 /* Poll infinite value of time */
975 ret
= lttng_poll_wait(&events
, -1);
976 DBG("Thread kernel return from poll on %d fds",
977 LTTNG_POLL_GETNB(&events
));
981 * Restart interrupted system call.
983 if (errno
== EINTR
) {
987 } else if (ret
== 0) {
988 /* Should not happen since timeout is infinite */
989 ERR("Return value of poll is 0 with an infinite timeout.\n"
990 "This should not have happened! Continuing...");
996 for (i
= 0; i
< nb_fd
; i
++) {
997 /* Fetch once the poll data */
998 revents
= LTTNG_POLL_GETEV(&events
, i
);
999 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1001 health_code_update();
1004 /* No activity for this FD (poll implementation). */
1008 /* Thread quit pipe has been closed. Killing thread. */
1009 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
1015 /* Check for data on kernel pipe */
1016 if (revents
& LPOLLIN
) {
1017 if (pollfd
== kernel_poll_pipe
[0]) {
1018 (void) lttng_read(kernel_poll_pipe
[0],
1021 * Ret value is useless here, if this pipe gets any actions an
1022 * update is required anyway.
1024 update_poll_flag
= 1;
1028 * New CPU detected by the kernel. Adding kernel stream to
1029 * kernel session and updating the kernel consumer
1031 ret
= update_kernel_stream(&kconsumer_data
, pollfd
);
1037 } else if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1038 update_poll_flag
= 1;
1041 ERR("Unexpected poll events %u for sock %d", revents
, pollfd
);
1049 lttng_poll_clean(&events
);
1052 utils_close_pipe(kernel_poll_pipe
);
1053 kernel_poll_pipe
[0] = kernel_poll_pipe
[1] = -1;
1056 ERR("Health error occurred in %s", __func__
);
1057 WARN("Kernel thread died unexpectedly. "
1058 "Kernel tracing can continue but CPU hotplug is disabled.");
1060 health_unregister(health_sessiond
);
1061 DBG("Kernel thread dying");
1066 * Signal pthread condition of the consumer data that the thread.
1068 static void signal_consumer_condition(struct consumer_data
*data
, int state
)
1070 pthread_mutex_lock(&data
->cond_mutex
);
1073 * The state is set before signaling. It can be any value, it's the waiter
1074 * job to correctly interpret this condition variable associated to the
1075 * consumer pthread_cond.
1077 * A value of 0 means that the corresponding thread of the consumer data
1078 * was not started. 1 indicates that the thread has started and is ready
1079 * for action. A negative value means that there was an error during the
1082 data
->consumer_thread_is_ready
= state
;
1083 (void) pthread_cond_signal(&data
->cond
);
1085 pthread_mutex_unlock(&data
->cond_mutex
);
1089 * This thread manage the consumer error sent back to the session daemon.
1091 static void *thread_manage_consumer(void *data
)
1093 int sock
= -1, i
, ret
, pollfd
, err
= -1, should_quit
= 0;
1094 uint32_t revents
, nb_fd
;
1095 enum lttcomm_return_code code
;
1096 struct lttng_poll_event events
;
1097 struct consumer_data
*consumer_data
= data
;
1098 struct consumer_socket
*cmd_socket_wrapper
= NULL
;
1100 DBG("[thread] Manage consumer started");
1102 rcu_register_thread();
1103 rcu_thread_online();
1105 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_CONSUMER
);
1107 health_code_update();
1110 * Pass 3 as size here for the thread quit pipe, consumerd_err_sock and the
1111 * metadata_sock. Nothing more will be added to this poll set.
1113 ret
= sessiond_set_thread_pollset(&events
, 3);
1119 * The error socket here is already in a listening state which was done
1120 * just before spawning this thread to avoid a race between the consumer
1121 * daemon exec trying to connect and the listen() call.
1123 ret
= lttng_poll_add(&events
, consumer_data
->err_sock
, LPOLLIN
| LPOLLRDHUP
);
1128 health_code_update();
1130 /* Infinite blocking call, waiting for transmission */
1132 health_poll_entry();
1134 if (testpoint(sessiond_thread_manage_consumer
)) {
1138 ret
= lttng_poll_wait(&events
, -1);
1142 * Restart interrupted system call.
1144 if (errno
== EINTR
) {
1152 for (i
= 0; i
< nb_fd
; i
++) {
1153 /* Fetch once the poll data */
1154 revents
= LTTNG_POLL_GETEV(&events
, i
);
1155 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1157 health_code_update();
1160 /* No activity for this FD (poll implementation). */
1164 /* Thread quit pipe has been closed. Killing thread. */
1165 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
1171 /* Event on the registration socket */
1172 if (pollfd
== consumer_data
->err_sock
) {
1173 if (revents
& LPOLLIN
) {
1175 } else if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1176 ERR("consumer err socket poll error");
1179 ERR("Unexpected poll events %u for sock %d", revents
, pollfd
);
1185 sock
= lttcomm_accept_unix_sock(consumer_data
->err_sock
);
1191 * Set the CLOEXEC flag. Return code is useless because either way, the
1194 (void) utils_set_fd_cloexec(sock
);
1196 health_code_update();
1198 DBG2("Receiving code from consumer err_sock");
1200 /* Getting status code from kconsumerd */
1201 ret
= lttcomm_recv_unix_sock(sock
, &code
,
1202 sizeof(enum lttcomm_return_code
));
1207 health_code_update();
1208 if (code
!= LTTCOMM_CONSUMERD_COMMAND_SOCK_READY
) {
1209 ERR("consumer error when waiting for SOCK_READY : %s",
1210 lttcomm_get_readable_code(-code
));
1214 /* Connect both command and metadata sockets. */
1215 consumer_data
->cmd_sock
=
1216 lttcomm_connect_unix_sock(
1217 consumer_data
->cmd_unix_sock_path
);
1218 consumer_data
->metadata_fd
=
1219 lttcomm_connect_unix_sock(
1220 consumer_data
->cmd_unix_sock_path
);
1221 if (consumer_data
->cmd_sock
< 0 || consumer_data
->metadata_fd
< 0) {
1222 PERROR("consumer connect cmd socket");
1223 /* On error, signal condition and quit. */
1224 signal_consumer_condition(consumer_data
, -1);
1228 consumer_data
->metadata_sock
.fd_ptr
= &consumer_data
->metadata_fd
;
1230 /* Create metadata socket lock. */
1231 consumer_data
->metadata_sock
.lock
= zmalloc(sizeof(pthread_mutex_t
));
1232 if (consumer_data
->metadata_sock
.lock
== NULL
) {
1233 PERROR("zmalloc pthread mutex");
1236 pthread_mutex_init(consumer_data
->metadata_sock
.lock
, NULL
);
1238 DBG("Consumer command socket ready (fd: %d", consumer_data
->cmd_sock
);
1239 DBG("Consumer metadata socket ready (fd: %d)",
1240 consumer_data
->metadata_fd
);
1243 * Remove the consumerd error sock since we've established a connection.
1245 ret
= lttng_poll_del(&events
, consumer_data
->err_sock
);
1250 /* Add new accepted error socket. */
1251 ret
= lttng_poll_add(&events
, sock
, LPOLLIN
| LPOLLRDHUP
);
1256 /* Add metadata socket that is successfully connected. */
1257 ret
= lttng_poll_add(&events
, consumer_data
->metadata_fd
,
1258 LPOLLIN
| LPOLLRDHUP
);
1263 health_code_update();
1266 * Transfer the write-end of the channel monitoring pipe to the
1267 * by issuing a SET_CHANNEL_MONITOR_PIPE command.
1269 cmd_socket_wrapper
= consumer_allocate_socket(&consumer_data
->cmd_sock
);
1270 if (!cmd_socket_wrapper
) {
1273 cmd_socket_wrapper
->lock
= &consumer_data
->lock
;
1275 ret
= consumer_send_channel_monitor_pipe(cmd_socket_wrapper
,
1276 consumer_data
->channel_monitor_pipe
);
1280 /* Discard the socket wrapper as it is no longer needed. */
1281 consumer_destroy_socket(cmd_socket_wrapper
);
1282 cmd_socket_wrapper
= NULL
;
1284 /* The thread is completely initialized, signal that it is ready. */
1285 signal_consumer_condition(consumer_data
, 1);
1287 /* Infinite blocking call, waiting for transmission */
1290 health_code_update();
1292 /* Exit the thread because the thread quit pipe has been triggered. */
1294 /* Not a health error. */
1299 health_poll_entry();
1300 ret
= lttng_poll_wait(&events
, -1);
1304 * Restart interrupted system call.
1306 if (errno
== EINTR
) {
1314 for (i
= 0; i
< nb_fd
; i
++) {
1315 /* Fetch once the poll data */
1316 revents
= LTTNG_POLL_GETEV(&events
, i
);
1317 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1319 health_code_update();
1322 /* No activity for this FD (poll implementation). */
1327 * Thread quit pipe has been triggered, flag that we should stop
1328 * but continue the current loop to handle potential data from
1331 should_quit
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
1333 if (pollfd
== sock
) {
1334 /* Event on the consumerd socket */
1335 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)
1336 && !(revents
& LPOLLIN
)) {
1337 ERR("consumer err socket second poll error");
1340 health_code_update();
1341 /* Wait for any kconsumerd error */
1342 ret
= lttcomm_recv_unix_sock(sock
, &code
,
1343 sizeof(enum lttcomm_return_code
));
1345 ERR("consumer closed the command socket");
1349 ERR("consumer return code : %s",
1350 lttcomm_get_readable_code(-code
));
1353 } else if (pollfd
== consumer_data
->metadata_fd
) {
1354 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)
1355 && !(revents
& LPOLLIN
)) {
1356 ERR("consumer err metadata socket second poll error");
1359 /* UST metadata requests */
1360 ret
= ust_consumer_metadata_request(
1361 &consumer_data
->metadata_sock
);
1363 ERR("Handling metadata request");
1367 /* No need for an else branch all FDs are tested prior. */
1369 health_code_update();
1375 * We lock here because we are about to close the sockets and some other
1376 * thread might be using them so get exclusive access which will abort all
1377 * other consumer command by other threads.
1379 pthread_mutex_lock(&consumer_data
->lock
);
1381 /* Immediately set the consumerd state to stopped */
1382 if (consumer_data
->type
== LTTNG_CONSUMER_KERNEL
) {
1383 uatomic_set(&kernel_consumerd_state
, CONSUMER_ERROR
);
1384 } else if (consumer_data
->type
== LTTNG_CONSUMER64_UST
||
1385 consumer_data
->type
== LTTNG_CONSUMER32_UST
) {
1386 uatomic_set(&ust_consumerd_state
, CONSUMER_ERROR
);
1388 /* Code flow error... */
1392 if (consumer_data
->err_sock
>= 0) {
1393 ret
= close(consumer_data
->err_sock
);
1397 consumer_data
->err_sock
= -1;
1399 if (consumer_data
->cmd_sock
>= 0) {
1400 ret
= close(consumer_data
->cmd_sock
);
1404 consumer_data
->cmd_sock
= -1;
1406 if (consumer_data
->metadata_sock
.fd_ptr
&&
1407 *consumer_data
->metadata_sock
.fd_ptr
>= 0) {
1408 ret
= close(*consumer_data
->metadata_sock
.fd_ptr
);
1420 unlink(consumer_data
->err_unix_sock_path
);
1421 unlink(consumer_data
->cmd_unix_sock_path
);
1422 pthread_mutex_unlock(&consumer_data
->lock
);
1424 /* Cleanup metadata socket mutex. */
1425 if (consumer_data
->metadata_sock
.lock
) {
1426 pthread_mutex_destroy(consumer_data
->metadata_sock
.lock
);
1427 free(consumer_data
->metadata_sock
.lock
);
1429 lttng_poll_clean(&events
);
1431 if (cmd_socket_wrapper
) {
1432 consumer_destroy_socket(cmd_socket_wrapper
);
1437 ERR("Health error occurred in %s", __func__
);
1439 health_unregister(health_sessiond
);
1440 DBG("consumer thread cleanup completed");
1442 rcu_thread_offline();
1443 rcu_unregister_thread();
1449 * This thread receives application command sockets (FDs) on the
1450 * apps_cmd_pipe and waits (polls) on them until they are closed
1451 * or an error occurs.
1453 * At that point, it flushes the data (tracing and metadata) associated
1454 * with this application and tears down ust app sessions and other
1455 * associated data structures through ust_app_unregister().
1457 * Note that this thread never sends commands to the applications
1458 * through the command sockets; it merely listens for hang-ups
1459 * and errors on those sockets and cleans-up as they occur.
1461 static void *thread_manage_apps(void *data
)
1463 int i
, ret
, pollfd
, err
= -1;
1465 uint32_t revents
, nb_fd
;
1466 struct lttng_poll_event events
;
1468 DBG("[thread] Manage application started");
1470 rcu_register_thread();
1471 rcu_thread_online();
1473 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_APP_MANAGE
);
1475 if (testpoint(sessiond_thread_manage_apps
)) {
1476 goto error_testpoint
;
1479 health_code_update();
1481 ret
= sessiond_set_thread_pollset(&events
, 2);
1483 goto error_poll_create
;
1486 ret
= lttng_poll_add(&events
, apps_cmd_pipe
[0], LPOLLIN
| LPOLLRDHUP
);
1491 if (testpoint(sessiond_thread_manage_apps_before_loop
)) {
1495 health_code_update();
1498 DBG("Apps thread polling");
1500 /* Inifinite blocking call, waiting for transmission */
1502 health_poll_entry();
1503 ret
= lttng_poll_wait(&events
, -1);
1504 DBG("Apps thread return from poll on %d fds",
1505 LTTNG_POLL_GETNB(&events
));
1509 * Restart interrupted system call.
1511 if (errno
== EINTR
) {
1519 for (i
= 0; i
< nb_fd
; i
++) {
1520 /* Fetch once the poll data */
1521 revents
= LTTNG_POLL_GETEV(&events
, i
);
1522 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1524 health_code_update();
1527 /* No activity for this FD (poll implementation). */
1531 /* Thread quit pipe has been closed. Killing thread. */
1532 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
1538 /* Inspect the apps cmd pipe */
1539 if (pollfd
== apps_cmd_pipe
[0]) {
1540 if (revents
& LPOLLIN
) {
1544 size_ret
= lttng_read(apps_cmd_pipe
[0], &sock
, sizeof(sock
));
1545 if (size_ret
< sizeof(sock
)) {
1546 PERROR("read apps cmd pipe");
1550 health_code_update();
1553 * Since this is a command socket (write then read),
1554 * we only monitor the error events of the socket.
1556 ret
= lttng_poll_add(&events
, sock
,
1557 LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
);
1562 DBG("Apps with sock %d added to poll set", sock
);
1563 } else if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1564 ERR("Apps command pipe error");
1567 ERR("Unknown poll events %u for sock %d", revents
, pollfd
);
1572 * At this point, we know that a registered application made
1573 * the event at poll_wait.
1575 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1576 /* Removing from the poll set */
1577 ret
= lttng_poll_del(&events
, pollfd
);
1582 /* Socket closed on remote end. */
1583 ust_app_unregister(pollfd
);
1585 ERR("Unexpected poll events %u for sock %d", revents
, pollfd
);
1590 health_code_update();
1596 lttng_poll_clean(&events
);
1599 utils_close_pipe(apps_cmd_pipe
);
1600 apps_cmd_pipe
[0] = apps_cmd_pipe
[1] = -1;
1603 * We don't clean the UST app hash table here since already registered
1604 * applications can still be controlled so let them be until the session
1605 * daemon dies or the applications stop.
1610 ERR("Health error occurred in %s", __func__
);
1612 health_unregister(health_sessiond
);
1613 DBG("Application communication apps thread cleanup complete");
1614 rcu_thread_offline();
1615 rcu_unregister_thread();
1620 * Send a socket to a thread This is called from the dispatch UST registration
1621 * thread once all sockets are set for the application.
1623 * The sock value can be invalid, we don't really care, the thread will handle
1624 * it and make the necessary cleanup if so.
1626 * On success, return 0 else a negative value being the errno message of the
1629 static int send_socket_to_thread(int fd
, int sock
)
1634 * It's possible that the FD is set as invalid with -1 concurrently just
1635 * before calling this function being a shutdown state of the thread.
1642 ret
= lttng_write(fd
, &sock
, sizeof(sock
));
1643 if (ret
< sizeof(sock
)) {
1644 PERROR("write apps pipe %d", fd
);
1651 /* All good. Don't send back the write positive ret value. */
1658 * Sanitize the wait queue of the dispatch registration thread meaning removing
1659 * invalid nodes from it. This is to avoid memory leaks for the case the UST
1660 * notify socket is never received.
1662 static void sanitize_wait_queue(struct ust_reg_wait_queue
*wait_queue
)
1664 int ret
, nb_fd
= 0, i
;
1665 unsigned int fd_added
= 0;
1666 struct lttng_poll_event events
;
1667 struct ust_reg_wait_node
*wait_node
= NULL
, *tmp_wait_node
;
1671 lttng_poll_init(&events
);
1673 /* Just skip everything for an empty queue. */
1674 if (!wait_queue
->count
) {
1678 ret
= lttng_poll_create(&events
, wait_queue
->count
, LTTNG_CLOEXEC
);
1683 cds_list_for_each_entry_safe(wait_node
, tmp_wait_node
,
1684 &wait_queue
->head
, head
) {
1685 assert(wait_node
->app
);
1686 ret
= lttng_poll_add(&events
, wait_node
->app
->sock
,
1687 LPOLLHUP
| LPOLLERR
);
1700 * Poll but don't block so we can quickly identify the faulty events and
1701 * clean them afterwards from the wait queue.
1703 ret
= lttng_poll_wait(&events
, 0);
1709 for (i
= 0; i
< nb_fd
; i
++) {
1710 /* Get faulty FD. */
1711 uint32_t revents
= LTTNG_POLL_GETEV(&events
, i
);
1712 int pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1715 /* No activity for this FD (poll implementation). */
1719 cds_list_for_each_entry_safe(wait_node
, tmp_wait_node
,
1720 &wait_queue
->head
, head
) {
1721 if (pollfd
== wait_node
->app
->sock
&&
1722 (revents
& (LPOLLHUP
| LPOLLERR
))) {
1723 cds_list_del(&wait_node
->head
);
1724 wait_queue
->count
--;
1725 ust_app_destroy(wait_node
->app
);
1728 * Silence warning of use-after-free in
1729 * cds_list_for_each_entry_safe which uses
1730 * __typeof__(*wait_node).
1735 ERR("Unexpected poll events %u for sock %d", revents
, pollfd
);
1742 DBG("Wait queue sanitized, %d node were cleaned up", nb_fd
);
1746 lttng_poll_clean(&events
);
1750 lttng_poll_clean(&events
);
1752 ERR("Unable to sanitize wait queue");
1757 * Dispatch request from the registration threads to the application
1758 * communication thread.
1760 static void *thread_dispatch_ust_registration(void *data
)
1763 struct cds_wfcq_node
*node
;
1764 struct ust_command
*ust_cmd
= NULL
;
1765 struct ust_reg_wait_node
*wait_node
= NULL
, *tmp_wait_node
;
1766 struct ust_reg_wait_queue wait_queue
= {
1770 rcu_register_thread();
1772 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_APP_REG_DISPATCH
);
1774 if (testpoint(sessiond_thread_app_reg_dispatch
)) {
1775 goto error_testpoint
;
1778 health_code_update();
1780 CDS_INIT_LIST_HEAD(&wait_queue
.head
);
1782 DBG("[thread] Dispatch UST command started");
1785 health_code_update();
1787 /* Atomically prepare the queue futex */
1788 futex_nto1_prepare(&ust_cmd_queue
.futex
);
1790 if (CMM_LOAD_SHARED(dispatch_thread_exit
)) {
1795 struct ust_app
*app
= NULL
;
1799 * Make sure we don't have node(s) that have hung up before receiving
1800 * the notify socket. This is to clean the list in order to avoid
1801 * memory leaks from notify socket that are never seen.
1803 sanitize_wait_queue(&wait_queue
);
1805 health_code_update();
1806 /* Dequeue command for registration */
1807 node
= cds_wfcq_dequeue_blocking(&ust_cmd_queue
.head
, &ust_cmd_queue
.tail
);
1809 DBG("Woken up but nothing in the UST command queue");
1810 /* Continue thread execution */
1814 ust_cmd
= caa_container_of(node
, struct ust_command
, node
);
1816 DBG("Dispatching UST registration pid:%d ppid:%d uid:%d"
1817 " gid:%d sock:%d name:%s (version %d.%d)",
1818 ust_cmd
->reg_msg
.pid
, ust_cmd
->reg_msg
.ppid
,
1819 ust_cmd
->reg_msg
.uid
, ust_cmd
->reg_msg
.gid
,
1820 ust_cmd
->sock
, ust_cmd
->reg_msg
.name
,
1821 ust_cmd
->reg_msg
.major
, ust_cmd
->reg_msg
.minor
);
1823 if (ust_cmd
->reg_msg
.type
== USTCTL_SOCKET_CMD
) {
1824 wait_node
= zmalloc(sizeof(*wait_node
));
1826 PERROR("zmalloc wait_node dispatch");
1827 ret
= close(ust_cmd
->sock
);
1829 PERROR("close ust sock dispatch %d", ust_cmd
->sock
);
1831 lttng_fd_put(LTTNG_FD_APPS
, 1);
1835 CDS_INIT_LIST_HEAD(&wait_node
->head
);
1837 /* Create application object if socket is CMD. */
1838 wait_node
->app
= ust_app_create(&ust_cmd
->reg_msg
,
1840 if (!wait_node
->app
) {
1841 ret
= close(ust_cmd
->sock
);
1843 PERROR("close ust sock dispatch %d", ust_cmd
->sock
);
1845 lttng_fd_put(LTTNG_FD_APPS
, 1);
1851 * Add application to the wait queue so we can set the notify
1852 * socket before putting this object in the global ht.
1854 cds_list_add(&wait_node
->head
, &wait_queue
.head
);
1859 * We have to continue here since we don't have the notify
1860 * socket and the application MUST be added to the hash table
1861 * only at that moment.
1866 * Look for the application in the local wait queue and set the
1867 * notify socket if found.
1869 cds_list_for_each_entry_safe(wait_node
, tmp_wait_node
,
1870 &wait_queue
.head
, head
) {
1871 health_code_update();
1872 if (wait_node
->app
->pid
== ust_cmd
->reg_msg
.pid
) {
1873 wait_node
->app
->notify_sock
= ust_cmd
->sock
;
1874 cds_list_del(&wait_node
->head
);
1876 app
= wait_node
->app
;
1878 DBG3("UST app notify socket %d is set", ust_cmd
->sock
);
1884 * With no application at this stage the received socket is
1885 * basically useless so close it before we free the cmd data
1886 * structure for good.
1889 ret
= close(ust_cmd
->sock
);
1891 PERROR("close ust sock dispatch %d", ust_cmd
->sock
);
1893 lttng_fd_put(LTTNG_FD_APPS
, 1);
1900 * @session_lock_list
1902 * Lock the global session list so from the register up to the
1903 * registration done message, no thread can see the application
1904 * and change its state.
1906 session_lock_list();
1910 * Add application to the global hash table. This needs to be
1911 * done before the update to the UST registry can locate the
1916 /* Set app version. This call will print an error if needed. */
1917 (void) ust_app_version(app
);
1919 /* Send notify socket through the notify pipe. */
1920 ret
= send_socket_to_thread(apps_cmd_notify_pipe
[1],
1924 session_unlock_list();
1926 * No notify thread, stop the UST tracing. However, this is
1927 * not an internal error of the this thread thus setting
1928 * the health error code to a normal exit.
1935 * Update newly registered application with the tracing
1936 * registry info already enabled information.
1938 update_ust_app(app
->sock
);
1941 * Don't care about return value. Let the manage apps threads
1942 * handle app unregistration upon socket close.
1944 (void) ust_app_register_done(app
);
1947 * Even if the application socket has been closed, send the app
1948 * to the thread and unregistration will take place at that
1951 ret
= send_socket_to_thread(apps_cmd_pipe
[1], app
->sock
);
1954 session_unlock_list();
1956 * No apps. thread, stop the UST tracing. However, this is
1957 * not an internal error of the this thread thus setting
1958 * the health error code to a normal exit.
1965 session_unlock_list();
1967 } while (node
!= NULL
);
1969 health_poll_entry();
1970 /* Futex wait on queue. Blocking call on futex() */
1971 futex_nto1_wait(&ust_cmd_queue
.futex
);
1974 /* Normal exit, no error */
1978 /* Clean up wait queue. */
1979 cds_list_for_each_entry_safe(wait_node
, tmp_wait_node
,
1980 &wait_queue
.head
, head
) {
1981 cds_list_del(&wait_node
->head
);
1986 /* Empty command queue. */
1988 /* Dequeue command for registration */
1989 node
= cds_wfcq_dequeue_blocking(&ust_cmd_queue
.head
, &ust_cmd_queue
.tail
);
1993 ust_cmd
= caa_container_of(node
, struct ust_command
, node
);
1994 ret
= close(ust_cmd
->sock
);
1996 PERROR("close ust sock exit dispatch %d", ust_cmd
->sock
);
1998 lttng_fd_put(LTTNG_FD_APPS
, 1);
2003 DBG("Dispatch thread dying");
2006 ERR("Health error occurred in %s", __func__
);
2008 health_unregister(health_sessiond
);
2009 rcu_unregister_thread();
2014 * This thread manage application registration.
2016 static void *thread_registration_apps(void *data
)
2018 int sock
= -1, i
, ret
, pollfd
, err
= -1;
2019 uint32_t revents
, nb_fd
;
2020 struct lttng_poll_event events
;
2022 * Get allocated in this thread, enqueued to a global queue, dequeued and
2023 * freed in the manage apps thread.
2025 struct ust_command
*ust_cmd
= NULL
;
2027 DBG("[thread] Manage application registration started");
2029 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_APP_REG
);
2031 if (testpoint(sessiond_thread_registration_apps
)) {
2032 goto error_testpoint
;
2035 ret
= lttcomm_listen_unix_sock(apps_sock
);
2041 * Pass 2 as size here for the thread quit pipe and apps socket. Nothing
2042 * more will be added to this poll set.
2044 ret
= sessiond_set_thread_pollset(&events
, 2);
2046 goto error_create_poll
;
2049 /* Add the application registration socket */
2050 ret
= lttng_poll_add(&events
, apps_sock
, LPOLLIN
| LPOLLRDHUP
);
2052 goto error_poll_add
;
2055 /* Notify all applications to register */
2056 ret
= notify_ust_apps(1);
2058 ERR("Failed to notify applications or create the wait shared memory.\n"
2059 "Execution continues but there might be problem for already\n"
2060 "running applications that wishes to register.");
2064 DBG("Accepting application registration");
2066 /* Inifinite blocking call, waiting for transmission */
2068 health_poll_entry();
2069 ret
= lttng_poll_wait(&events
, -1);
2073 * Restart interrupted system call.
2075 if (errno
== EINTR
) {
2083 for (i
= 0; i
< nb_fd
; i
++) {
2084 health_code_update();
2086 /* Fetch once the poll data */
2087 revents
= LTTNG_POLL_GETEV(&events
, i
);
2088 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
2091 /* No activity for this FD (poll implementation). */
2095 /* Thread quit pipe has been closed. Killing thread. */
2096 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
2102 /* Event on the registration socket */
2103 if (pollfd
== apps_sock
) {
2104 if (revents
& LPOLLIN
) {
2105 sock
= lttcomm_accept_unix_sock(apps_sock
);
2111 * Set socket timeout for both receiving and ending.
2112 * app_socket_timeout is in seconds, whereas
2113 * lttcomm_setsockopt_rcv_timeout and
2114 * lttcomm_setsockopt_snd_timeout expect msec as
2117 if (config
.app_socket_timeout
>= 0) {
2118 (void) lttcomm_setsockopt_rcv_timeout(sock
,
2119 config
.app_socket_timeout
* 1000);
2120 (void) lttcomm_setsockopt_snd_timeout(sock
,
2121 config
.app_socket_timeout
* 1000);
2125 * Set the CLOEXEC flag. Return code is useless because
2126 * either way, the show must go on.
2128 (void) utils_set_fd_cloexec(sock
);
2130 /* Create UST registration command for enqueuing */
2131 ust_cmd
= zmalloc(sizeof(struct ust_command
));
2132 if (ust_cmd
== NULL
) {
2133 PERROR("ust command zmalloc");
2142 * Using message-based transmissions to ensure we don't
2143 * have to deal with partially received messages.
2145 ret
= lttng_fd_get(LTTNG_FD_APPS
, 1);
2147 ERR("Exhausted file descriptors allowed for applications.");
2157 health_code_update();
2158 ret
= ust_app_recv_registration(sock
, &ust_cmd
->reg_msg
);
2161 /* Close socket of the application. */
2166 lttng_fd_put(LTTNG_FD_APPS
, 1);
2170 health_code_update();
2172 ust_cmd
->sock
= sock
;
2175 DBG("UST registration received with pid:%d ppid:%d uid:%d"
2176 " gid:%d sock:%d name:%s (version %d.%d)",
2177 ust_cmd
->reg_msg
.pid
, ust_cmd
->reg_msg
.ppid
,
2178 ust_cmd
->reg_msg
.uid
, ust_cmd
->reg_msg
.gid
,
2179 ust_cmd
->sock
, ust_cmd
->reg_msg
.name
,
2180 ust_cmd
->reg_msg
.major
, ust_cmd
->reg_msg
.minor
);
2183 * Lock free enqueue the registration request. The red pill
2184 * has been taken! This apps will be part of the *system*.
2186 cds_wfcq_enqueue(&ust_cmd_queue
.head
, &ust_cmd_queue
.tail
, &ust_cmd
->node
);
2189 * Wake the registration queue futex. Implicit memory
2190 * barrier with the exchange in cds_wfcq_enqueue.
2192 futex_nto1_wake(&ust_cmd_queue
.futex
);
2193 } else if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
2194 ERR("Register apps socket poll error");
2197 ERR("Unexpected poll events %u for sock %d", revents
, pollfd
);
2206 /* Notify that the registration thread is gone */
2209 if (apps_sock
>= 0) {
2210 ret
= close(apps_sock
);
2220 lttng_fd_put(LTTNG_FD_APPS
, 1);
2222 unlink(config
.apps_unix_sock_path
.value
);
2225 lttng_poll_clean(&events
);
2229 DBG("UST Registration thread cleanup complete");
2232 ERR("Health error occurred in %s", __func__
);
2234 health_unregister(health_sessiond
);
2240 * Start the thread_manage_consumer. This must be done after a lttng-consumerd
2241 * exec or it will fails.
2243 static int spawn_consumer_thread(struct consumer_data
*consumer_data
)
2246 struct timespec timeout
;
2249 * Make sure we set the readiness flag to 0 because we are NOT ready.
2250 * This access to consumer_thread_is_ready does not need to be
2251 * protected by consumer_data.cond_mutex (yet) since the consumer
2252 * management thread has not been started at this point.
2254 consumer_data
->consumer_thread_is_ready
= 0;
2256 /* Setup pthread condition */
2257 ret
= pthread_condattr_init(&consumer_data
->condattr
);
2260 PERROR("pthread_condattr_init consumer data");
2265 * Set the monotonic clock in order to make sure we DO NOT jump in time
2266 * between the clock_gettime() call and the timedwait call. See bug #324
2267 * for a more details and how we noticed it.
2269 ret
= pthread_condattr_setclock(&consumer_data
->condattr
, CLOCK_MONOTONIC
);
2272 PERROR("pthread_condattr_setclock consumer data");
2276 ret
= pthread_cond_init(&consumer_data
->cond
, &consumer_data
->condattr
);
2279 PERROR("pthread_cond_init consumer data");
2283 ret
= pthread_create(&consumer_data
->thread
, default_pthread_attr(),
2284 thread_manage_consumer
, consumer_data
);
2287 PERROR("pthread_create consumer");
2292 /* We are about to wait on a pthread condition */
2293 pthread_mutex_lock(&consumer_data
->cond_mutex
);
2295 /* Get time for sem_timedwait absolute timeout */
2296 clock_ret
= lttng_clock_gettime(CLOCK_MONOTONIC
, &timeout
);
2298 * Set the timeout for the condition timed wait even if the clock gettime
2299 * call fails since we might loop on that call and we want to avoid to
2300 * increment the timeout too many times.
2302 timeout
.tv_sec
+= DEFAULT_SEM_WAIT_TIMEOUT
;
2305 * The following loop COULD be skipped in some conditions so this is why we
2306 * set ret to 0 in order to make sure at least one round of the loop is
2312 * Loop until the condition is reached or when a timeout is reached. Note
2313 * that the pthread_cond_timedwait(P) man page specifies that EINTR can NOT
2314 * be returned but the pthread_cond(3), from the glibc-doc, says that it is
2315 * possible. This loop does not take any chances and works with both of
2318 while (!consumer_data
->consumer_thread_is_ready
&& ret
!= ETIMEDOUT
) {
2319 if (clock_ret
< 0) {
2320 PERROR("clock_gettime spawn consumer");
2321 /* Infinite wait for the consumerd thread to be ready */
2322 ret
= pthread_cond_wait(&consumer_data
->cond
,
2323 &consumer_data
->cond_mutex
);
2325 ret
= pthread_cond_timedwait(&consumer_data
->cond
,
2326 &consumer_data
->cond_mutex
, &timeout
);
2330 /* Release the pthread condition */
2331 pthread_mutex_unlock(&consumer_data
->cond_mutex
);
2335 if (ret
== ETIMEDOUT
) {
2339 * Call has timed out so we kill the kconsumerd_thread and return
2342 ERR("Condition timed out. The consumer thread was never ready."
2344 pth_ret
= pthread_cancel(consumer_data
->thread
);
2346 PERROR("pthread_cancel consumer thread");
2349 PERROR("pthread_cond_wait failed consumer thread");
2351 /* Caller is expecting a negative value on failure. */
2356 pthread_mutex_lock(&consumer_data
->pid_mutex
);
2357 if (consumer_data
->pid
== 0) {
2358 ERR("Consumerd did not start");
2359 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2362 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2371 * Join consumer thread
2373 static int join_consumer_thread(struct consumer_data
*consumer_data
)
2377 /* Consumer pid must be a real one. */
2378 if (consumer_data
->pid
> 0) {
2380 ret
= kill(consumer_data
->pid
, SIGTERM
);
2382 PERROR("Error killing consumer daemon");
2385 return pthread_join(consumer_data
->thread
, &status
);
2392 * Fork and exec a consumer daemon (consumerd).
2394 * Return pid if successful else -1.
2396 static pid_t
spawn_consumerd(struct consumer_data
*consumer_data
)
2400 const char *consumer_to_use
;
2401 const char *verbosity
;
2404 DBG("Spawning consumerd");
2411 if (config
.verbose_consumer
) {
2412 verbosity
= "--verbose";
2413 } else if (lttng_opt_quiet
) {
2414 verbosity
= "--quiet";
2419 switch (consumer_data
->type
) {
2420 case LTTNG_CONSUMER_KERNEL
:
2422 * Find out which consumerd to execute. We will first try the
2423 * 64-bit path, then the sessiond's installation directory, and
2424 * fallback on the 32-bit one,
2426 DBG3("Looking for a kernel consumer at these locations:");
2427 DBG3(" 1) %s", config
.consumerd64_bin_path
.value
? : "NULL");
2428 DBG3(" 2) %s/%s", INSTALL_BIN_PATH
, DEFAULT_CONSUMERD_FILE
);
2429 DBG3(" 3) %s", config
.consumerd32_bin_path
.value
? : "NULL");
2430 if (stat(config
.consumerd64_bin_path
.value
, &st
) == 0) {
2431 DBG3("Found location #1");
2432 consumer_to_use
= config
.consumerd64_bin_path
.value
;
2433 } else if (stat(INSTALL_BIN_PATH
"/" DEFAULT_CONSUMERD_FILE
, &st
) == 0) {
2434 DBG3("Found location #2");
2435 consumer_to_use
= INSTALL_BIN_PATH
"/" DEFAULT_CONSUMERD_FILE
;
2436 } else if (stat(config
.consumerd32_bin_path
.value
, &st
) == 0) {
2437 DBG3("Found location #3");
2438 consumer_to_use
= config
.consumerd32_bin_path
.value
;
2440 DBG("Could not find any valid consumerd executable");
2444 DBG("Using kernel consumer at: %s", consumer_to_use
);
2445 (void) execl(consumer_to_use
,
2446 "lttng-consumerd", verbosity
, "-k",
2447 "--consumerd-cmd-sock", consumer_data
->cmd_unix_sock_path
,
2448 "--consumerd-err-sock", consumer_data
->err_unix_sock_path
,
2449 "--group", config
.tracing_group_name
.value
,
2452 case LTTNG_CONSUMER64_UST
:
2454 if (config
.consumerd64_lib_dir
.value
) {
2459 tmp
= lttng_secure_getenv("LD_LIBRARY_PATH");
2463 tmplen
= strlen(config
.consumerd64_lib_dir
.value
) + 1 /* : */ + strlen(tmp
);
2464 tmpnew
= zmalloc(tmplen
+ 1 /* \0 */);
2469 strcat(tmpnew
, config
.consumerd64_lib_dir
.value
);
2470 if (tmp
[0] != '\0') {
2471 strcat(tmpnew
, ":");
2472 strcat(tmpnew
, tmp
);
2474 ret
= setenv("LD_LIBRARY_PATH", tmpnew
, 1);
2481 DBG("Using 64-bit UST consumer at: %s", config
.consumerd64_bin_path
.value
);
2482 (void) execl(config
.consumerd64_bin_path
.value
, "lttng-consumerd", verbosity
, "-u",
2483 "--consumerd-cmd-sock", consumer_data
->cmd_unix_sock_path
,
2484 "--consumerd-err-sock", consumer_data
->err_unix_sock_path
,
2485 "--group", config
.tracing_group_name
.value
,
2489 case LTTNG_CONSUMER32_UST
:
2491 if (config
.consumerd32_lib_dir
.value
) {
2496 tmp
= lttng_secure_getenv("LD_LIBRARY_PATH");
2500 tmplen
= strlen(config
.consumerd32_lib_dir
.value
) + 1 /* : */ + strlen(tmp
);
2501 tmpnew
= zmalloc(tmplen
+ 1 /* \0 */);
2506 strcat(tmpnew
, config
.consumerd32_lib_dir
.value
);
2507 if (tmp
[0] != '\0') {
2508 strcat(tmpnew
, ":");
2509 strcat(tmpnew
, tmp
);
2511 ret
= setenv("LD_LIBRARY_PATH", tmpnew
, 1);
2518 DBG("Using 32-bit UST consumer at: %s", config
.consumerd32_bin_path
.value
);
2519 (void) execl(config
.consumerd32_bin_path
.value
, "lttng-consumerd", verbosity
, "-u",
2520 "--consumerd-cmd-sock", consumer_data
->cmd_unix_sock_path
,
2521 "--consumerd-err-sock", consumer_data
->err_unix_sock_path
,
2522 "--group", config
.tracing_group_name
.value
,
2527 ERR("unknown consumer type");
2531 PERROR("Consumer execl()");
2533 /* Reaching this point, we got a failure on our execl(). */
2535 } else if (pid
> 0) {
2538 PERROR("start consumer fork");
2546 * Spawn the consumerd daemon and session daemon thread.
2548 static int start_consumerd(struct consumer_data
*consumer_data
)
2553 * Set the listen() state on the socket since there is a possible race
2554 * between the exec() of the consumer daemon and this call if place in the
2555 * consumer thread. See bug #366 for more details.
2557 ret
= lttcomm_listen_unix_sock(consumer_data
->err_sock
);
2562 pthread_mutex_lock(&consumer_data
->pid_mutex
);
2563 if (consumer_data
->pid
!= 0) {
2564 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2568 ret
= spawn_consumerd(consumer_data
);
2570 ERR("Spawning consumerd failed");
2571 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2575 /* Setting up the consumer_data pid */
2576 consumer_data
->pid
= ret
;
2577 DBG2("Consumer pid %d", consumer_data
->pid
);
2578 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2580 DBG2("Spawning consumer control thread");
2581 ret
= spawn_consumer_thread(consumer_data
);
2583 ERR("Fatal error spawning consumer control thread");
2591 /* Cleanup already created sockets on error. */
2592 if (consumer_data
->err_sock
>= 0) {
2595 err
= close(consumer_data
->err_sock
);
2597 PERROR("close consumer data error socket");
2604 * Setup necessary data for kernel tracer action.
2606 static int init_kernel_tracer(void)
2610 /* Modprobe lttng kernel modules */
2611 ret
= modprobe_lttng_control();
2616 /* Open debugfs lttng */
2617 kernel_tracer_fd
= open(module_proc_lttng
, O_RDWR
);
2618 if (kernel_tracer_fd
< 0) {
2619 DBG("Failed to open %s", module_proc_lttng
);
2623 /* Validate kernel version */
2624 ret
= kernel_validate_version(kernel_tracer_fd
);
2629 ret
= modprobe_lttng_data();
2634 ret
= kernel_supports_ring_buffer_snapshot_sample_positions(
2641 WARN("Kernel tracer does not support buffer monitoring. "
2642 "The monitoring timer of channels in the kernel domain "
2643 "will be set to 0 (disabled).");
2646 DBG("Kernel tracer fd %d", kernel_tracer_fd
);
2650 modprobe_remove_lttng_control();
2651 ret
= close(kernel_tracer_fd
);
2655 kernel_tracer_fd
= -1;
2656 return LTTNG_ERR_KERN_VERSION
;
2659 ret
= close(kernel_tracer_fd
);
2665 modprobe_remove_lttng_control();
2668 WARN("No kernel tracer available");
2669 kernel_tracer_fd
= -1;
2671 return LTTNG_ERR_NEED_ROOT_SESSIOND
;
2673 return LTTNG_ERR_KERN_NA
;
2679 * Copy consumer output from the tracing session to the domain session. The
2680 * function also applies the right modification on a per domain basis for the
2681 * trace files destination directory.
2683 * Should *NOT* be called with RCU read-side lock held.
2685 static int copy_session_consumer(int domain
, struct ltt_session
*session
)
2688 const char *dir_name
;
2689 struct consumer_output
*consumer
;
2692 assert(session
->consumer
);
2695 case LTTNG_DOMAIN_KERNEL
:
2696 DBG3("Copying tracing session consumer output in kernel session");
2698 * XXX: We should audit the session creation and what this function
2699 * does "extra" in order to avoid a destroy since this function is used
2700 * in the domain session creation (kernel and ust) only. Same for UST
2703 if (session
->kernel_session
->consumer
) {
2704 consumer_output_put(session
->kernel_session
->consumer
);
2706 session
->kernel_session
->consumer
=
2707 consumer_copy_output(session
->consumer
);
2708 /* Ease our life a bit for the next part */
2709 consumer
= session
->kernel_session
->consumer
;
2710 dir_name
= DEFAULT_KERNEL_TRACE_DIR
;
2712 case LTTNG_DOMAIN_JUL
:
2713 case LTTNG_DOMAIN_LOG4J
:
2714 case LTTNG_DOMAIN_PYTHON
:
2715 case LTTNG_DOMAIN_UST
:
2716 DBG3("Copying tracing session consumer output in UST session");
2717 if (session
->ust_session
->consumer
) {
2718 consumer_output_put(session
->ust_session
->consumer
);
2720 session
->ust_session
->consumer
=
2721 consumer_copy_output(session
->consumer
);
2722 /* Ease our life a bit for the next part */
2723 consumer
= session
->ust_session
->consumer
;
2724 dir_name
= DEFAULT_UST_TRACE_DIR
;
2727 ret
= LTTNG_ERR_UNKNOWN_DOMAIN
;
2731 /* Append correct directory to subdir */
2732 strncat(consumer
->subdir
, dir_name
,
2733 sizeof(consumer
->subdir
) - strlen(consumer
->subdir
) - 1);
2734 DBG3("Copy session consumer subdir %s", consumer
->subdir
);
2743 * Create an UST session and add it to the session ust list.
2745 * Should *NOT* be called with RCU read-side lock held.
2747 static int create_ust_session(struct ltt_session
*session
,
2748 struct lttng_domain
*domain
)
2751 struct ltt_ust_session
*lus
= NULL
;
2755 assert(session
->consumer
);
2757 switch (domain
->type
) {
2758 case LTTNG_DOMAIN_JUL
:
2759 case LTTNG_DOMAIN_LOG4J
:
2760 case LTTNG_DOMAIN_PYTHON
:
2761 case LTTNG_DOMAIN_UST
:
2764 ERR("Unknown UST domain on create session %d", domain
->type
);
2765 ret
= LTTNG_ERR_UNKNOWN_DOMAIN
;
2769 DBG("Creating UST session");
2771 lus
= trace_ust_create_session(session
->id
);
2773 ret
= LTTNG_ERR_UST_SESS_FAIL
;
2777 lus
->uid
= session
->uid
;
2778 lus
->gid
= session
->gid
;
2779 lus
->output_traces
= session
->output_traces
;
2780 lus
->snapshot_mode
= session
->snapshot_mode
;
2781 lus
->live_timer_interval
= session
->live_timer
;
2782 session
->ust_session
= lus
;
2783 if (session
->shm_path
[0]) {
2784 strncpy(lus
->root_shm_path
, session
->shm_path
,
2785 sizeof(lus
->root_shm_path
));
2786 lus
->root_shm_path
[sizeof(lus
->root_shm_path
) - 1] = '\0';
2787 strncpy(lus
->shm_path
, session
->shm_path
,
2788 sizeof(lus
->shm_path
));
2789 lus
->shm_path
[sizeof(lus
->shm_path
) - 1] = '\0';
2790 strncat(lus
->shm_path
, "/ust",
2791 sizeof(lus
->shm_path
) - strlen(lus
->shm_path
) - 1);
2793 /* Copy session output to the newly created UST session */
2794 ret
= copy_session_consumer(domain
->type
, session
);
2795 if (ret
!= LTTNG_OK
) {
2803 session
->ust_session
= NULL
;
2808 * Create a kernel tracer session then create the default channel.
2810 static int create_kernel_session(struct ltt_session
*session
)
2814 DBG("Creating kernel session");
2816 ret
= kernel_create_session(session
, kernel_tracer_fd
);
2818 ret
= LTTNG_ERR_KERN_SESS_FAIL
;
2822 /* Code flow safety */
2823 assert(session
->kernel_session
);
2825 /* Copy session output to the newly created Kernel session */
2826 ret
= copy_session_consumer(LTTNG_DOMAIN_KERNEL
, session
);
2827 if (ret
!= LTTNG_OK
) {
2831 /* Create directory(ies) on local filesystem. */
2832 if (session
->kernel_session
->consumer
->type
== CONSUMER_DST_LOCAL
&&
2833 strlen(session
->kernel_session
->consumer
->dst
.trace_path
) > 0) {
2834 ret
= run_as_mkdir_recursive(
2835 session
->kernel_session
->consumer
->dst
.trace_path
,
2836 S_IRWXU
| S_IRWXG
, session
->uid
, session
->gid
);
2838 if (errno
!= EEXIST
) {
2839 ERR("Trace directory creation error");
2845 session
->kernel_session
->uid
= session
->uid
;
2846 session
->kernel_session
->gid
= session
->gid
;
2847 session
->kernel_session
->output_traces
= session
->output_traces
;
2848 session
->kernel_session
->snapshot_mode
= session
->snapshot_mode
;
2853 trace_kernel_destroy_session(session
->kernel_session
);
2854 session
->kernel_session
= NULL
;
2859 * Count number of session permitted by uid/gid.
2861 static unsigned int lttng_sessions_count(uid_t uid
, gid_t gid
)
2864 struct ltt_session
*session
;
2866 DBG("Counting number of available session for UID %d GID %d",
2868 cds_list_for_each_entry(session
, &session_list_ptr
->head
, list
) {
2870 * Only list the sessions the user can control.
2872 if (!session_access_ok(session
, uid
, gid
)) {
2881 * Process the command requested by the lttng client within the command
2882 * context structure. This function make sure that the return structure (llm)
2883 * is set and ready for transmission before returning.
2885 * Return any error encountered or 0 for success.
2887 * "sock" is only used for special-case var. len data.
2889 * Should *NOT* be called with RCU read-side lock held.
2891 static int process_client_msg(struct command_ctx
*cmd_ctx
, int sock
,
2895 int need_tracing_session
= 1;
2898 DBG("Processing client command %d", cmd_ctx
->lsm
->cmd_type
);
2900 assert(!rcu_read_ongoing());
2904 switch (cmd_ctx
->lsm
->cmd_type
) {
2905 case LTTNG_CREATE_SESSION
:
2906 case LTTNG_CREATE_SESSION_SNAPSHOT
:
2907 case LTTNG_CREATE_SESSION_LIVE
:
2908 case LTTNG_DESTROY_SESSION
:
2909 case LTTNG_LIST_SESSIONS
:
2910 case LTTNG_LIST_DOMAINS
:
2911 case LTTNG_START_TRACE
:
2912 case LTTNG_STOP_TRACE
:
2913 case LTTNG_DATA_PENDING
:
2914 case LTTNG_SNAPSHOT_ADD_OUTPUT
:
2915 case LTTNG_SNAPSHOT_DEL_OUTPUT
:
2916 case LTTNG_SNAPSHOT_LIST_OUTPUT
:
2917 case LTTNG_SNAPSHOT_RECORD
:
2918 case LTTNG_SAVE_SESSION
:
2919 case LTTNG_SET_SESSION_SHM_PATH
:
2920 case LTTNG_REGENERATE_METADATA
:
2921 case LTTNG_REGENERATE_STATEDUMP
:
2922 case LTTNG_REGISTER_TRIGGER
:
2923 case LTTNG_UNREGISTER_TRIGGER
:
2930 if (config
.no_kernel
&& need_domain
2931 && cmd_ctx
->lsm
->domain
.type
== LTTNG_DOMAIN_KERNEL
) {
2933 ret
= LTTNG_ERR_NEED_ROOT_SESSIOND
;
2935 ret
= LTTNG_ERR_KERN_NA
;
2940 /* Deny register consumer if we already have a spawned consumer. */
2941 if (cmd_ctx
->lsm
->cmd_type
== LTTNG_REGISTER_CONSUMER
) {
2942 pthread_mutex_lock(&kconsumer_data
.pid_mutex
);
2943 if (kconsumer_data
.pid
> 0) {
2944 ret
= LTTNG_ERR_KERN_CONSUMER_FAIL
;
2945 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
2948 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
2952 * Check for command that don't needs to allocate a returned payload. We do
2953 * this here so we don't have to make the call for no payload at each
2956 switch(cmd_ctx
->lsm
->cmd_type
) {
2957 case LTTNG_LIST_SESSIONS
:
2958 case LTTNG_LIST_TRACEPOINTS
:
2959 case LTTNG_LIST_TRACEPOINT_FIELDS
:
2960 case LTTNG_LIST_DOMAINS
:
2961 case LTTNG_LIST_CHANNELS
:
2962 case LTTNG_LIST_EVENTS
:
2963 case LTTNG_LIST_SYSCALLS
:
2964 case LTTNG_LIST_TRACKER_PIDS
:
2965 case LTTNG_DATA_PENDING
:
2968 /* Setup lttng message with no payload */
2969 ret
= setup_lttng_msg_no_cmd_header(cmd_ctx
, NULL
, 0);
2971 /* This label does not try to unlock the session */
2972 goto init_setup_error
;
2976 /* Commands that DO NOT need a session. */
2977 switch (cmd_ctx
->lsm
->cmd_type
) {
2978 case LTTNG_CREATE_SESSION
:
2979 case LTTNG_CREATE_SESSION_SNAPSHOT
:
2980 case LTTNG_CREATE_SESSION_LIVE
:
2981 case LTTNG_LIST_SESSIONS
:
2982 case LTTNG_LIST_TRACEPOINTS
:
2983 case LTTNG_LIST_SYSCALLS
:
2984 case LTTNG_LIST_TRACEPOINT_FIELDS
:
2985 case LTTNG_SAVE_SESSION
:
2986 case LTTNG_REGISTER_TRIGGER
:
2987 case LTTNG_UNREGISTER_TRIGGER
:
2988 need_tracing_session
= 0;
2991 DBG("Getting session %s by name", cmd_ctx
->lsm
->session
.name
);
2993 * We keep the session list lock across _all_ commands
2994 * for now, because the per-session lock does not
2995 * handle teardown properly.
2997 session_lock_list();
2998 cmd_ctx
->session
= session_find_by_name(cmd_ctx
->lsm
->session
.name
);
2999 if (cmd_ctx
->session
== NULL
) {
3000 ret
= LTTNG_ERR_SESS_NOT_FOUND
;
3003 /* Acquire lock for the session */
3004 session_lock(cmd_ctx
->session
);
3010 * Commands that need a valid session but should NOT create one if none
3011 * exists. Instead of creating one and destroying it when the command is
3012 * handled, process that right before so we save some round trip in useless
3015 switch (cmd_ctx
->lsm
->cmd_type
) {
3016 case LTTNG_DISABLE_CHANNEL
:
3017 case LTTNG_DISABLE_EVENT
:
3018 switch (cmd_ctx
->lsm
->domain
.type
) {
3019 case LTTNG_DOMAIN_KERNEL
:
3020 if (!cmd_ctx
->session
->kernel_session
) {
3021 ret
= LTTNG_ERR_NO_CHANNEL
;
3025 case LTTNG_DOMAIN_JUL
:
3026 case LTTNG_DOMAIN_LOG4J
:
3027 case LTTNG_DOMAIN_PYTHON
:
3028 case LTTNG_DOMAIN_UST
:
3029 if (!cmd_ctx
->session
->ust_session
) {
3030 ret
= LTTNG_ERR_NO_CHANNEL
;
3035 ret
= LTTNG_ERR_UNKNOWN_DOMAIN
;
3047 * Check domain type for specific "pre-action".
3049 switch (cmd_ctx
->lsm
->domain
.type
) {
3050 case LTTNG_DOMAIN_KERNEL
:
3052 ret
= LTTNG_ERR_NEED_ROOT_SESSIOND
;
3056 /* Kernel tracer check */
3057 if (kernel_tracer_fd
== -1) {
3058 /* Basically, load kernel tracer modules */
3059 ret
= init_kernel_tracer();
3065 /* Consumer is in an ERROR state. Report back to client */
3066 if (uatomic_read(&kernel_consumerd_state
) == CONSUMER_ERROR
) {
3067 ret
= LTTNG_ERR_NO_KERNCONSUMERD
;
3071 /* Need a session for kernel command */
3072 if (need_tracing_session
) {
3073 if (cmd_ctx
->session
->kernel_session
== NULL
) {
3074 ret
= create_kernel_session(cmd_ctx
->session
);
3076 ret
= LTTNG_ERR_KERN_SESS_FAIL
;
3081 /* Start the kernel consumer daemon */
3082 pthread_mutex_lock(&kconsumer_data
.pid_mutex
);
3083 if (kconsumer_data
.pid
== 0 &&
3084 cmd_ctx
->lsm
->cmd_type
!= LTTNG_REGISTER_CONSUMER
) {
3085 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
3086 ret
= start_consumerd(&kconsumer_data
);
3088 ret
= LTTNG_ERR_KERN_CONSUMER_FAIL
;
3091 uatomic_set(&kernel_consumerd_state
, CONSUMER_STARTED
);
3093 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
3097 * The consumer was just spawned so we need to add the socket to
3098 * the consumer output of the session if exist.
3100 ret
= consumer_create_socket(&kconsumer_data
,
3101 cmd_ctx
->session
->kernel_session
->consumer
);
3108 case LTTNG_DOMAIN_JUL
:
3109 case LTTNG_DOMAIN_LOG4J
:
3110 case LTTNG_DOMAIN_PYTHON
:
3111 case LTTNG_DOMAIN_UST
:
3113 if (!ust_app_supported()) {
3114 ret
= LTTNG_ERR_NO_UST
;
3117 /* Consumer is in an ERROR state. Report back to client */
3118 if (uatomic_read(&ust_consumerd_state
) == CONSUMER_ERROR
) {
3119 ret
= LTTNG_ERR_NO_USTCONSUMERD
;
3123 if (need_tracing_session
) {
3124 /* Create UST session if none exist. */
3125 if (cmd_ctx
->session
->ust_session
== NULL
) {
3126 ret
= create_ust_session(cmd_ctx
->session
,
3127 &cmd_ctx
->lsm
->domain
);
3128 if (ret
!= LTTNG_OK
) {
3133 /* Start the UST consumer daemons */
3135 pthread_mutex_lock(&ustconsumer64_data
.pid_mutex
);
3136 if (config
.consumerd64_bin_path
.value
&&
3137 ustconsumer64_data
.pid
== 0 &&
3138 cmd_ctx
->lsm
->cmd_type
!= LTTNG_REGISTER_CONSUMER
) {
3139 pthread_mutex_unlock(&ustconsumer64_data
.pid_mutex
);
3140 ret
= start_consumerd(&ustconsumer64_data
);
3142 ret
= LTTNG_ERR_UST_CONSUMER64_FAIL
;
3143 uatomic_set(&ust_consumerd64_fd
, -EINVAL
);
3147 uatomic_set(&ust_consumerd64_fd
, ustconsumer64_data
.cmd_sock
);
3148 uatomic_set(&ust_consumerd_state
, CONSUMER_STARTED
);
3150 pthread_mutex_unlock(&ustconsumer64_data
.pid_mutex
);
3154 * Setup socket for consumer 64 bit. No need for atomic access
3155 * since it was set above and can ONLY be set in this thread.
3157 ret
= consumer_create_socket(&ustconsumer64_data
,
3158 cmd_ctx
->session
->ust_session
->consumer
);
3164 pthread_mutex_lock(&ustconsumer32_data
.pid_mutex
);
3165 if (config
.consumerd32_bin_path
.value
&&
3166 ustconsumer32_data
.pid
== 0 &&
3167 cmd_ctx
->lsm
->cmd_type
!= LTTNG_REGISTER_CONSUMER
) {
3168 pthread_mutex_unlock(&ustconsumer32_data
.pid_mutex
);
3169 ret
= start_consumerd(&ustconsumer32_data
);
3171 ret
= LTTNG_ERR_UST_CONSUMER32_FAIL
;
3172 uatomic_set(&ust_consumerd32_fd
, -EINVAL
);
3176 uatomic_set(&ust_consumerd32_fd
, ustconsumer32_data
.cmd_sock
);
3177 uatomic_set(&ust_consumerd_state
, CONSUMER_STARTED
);
3179 pthread_mutex_unlock(&ustconsumer32_data
.pid_mutex
);
3183 * Setup socket for consumer 64 bit. No need for atomic access
3184 * since it was set above and can ONLY be set in this thread.
3186 ret
= consumer_create_socket(&ustconsumer32_data
,
3187 cmd_ctx
->session
->ust_session
->consumer
);
3199 /* Validate consumer daemon state when start/stop trace command */
3200 if (cmd_ctx
->lsm
->cmd_type
== LTTNG_START_TRACE
||
3201 cmd_ctx
->lsm
->cmd_type
== LTTNG_STOP_TRACE
) {
3202 switch (cmd_ctx
->lsm
->domain
.type
) {
3203 case LTTNG_DOMAIN_NONE
:
3205 case LTTNG_DOMAIN_JUL
:
3206 case LTTNG_DOMAIN_LOG4J
:
3207 case LTTNG_DOMAIN_PYTHON
:
3208 case LTTNG_DOMAIN_UST
:
3209 if (uatomic_read(&ust_consumerd_state
) != CONSUMER_STARTED
) {
3210 ret
= LTTNG_ERR_NO_USTCONSUMERD
;
3214 case LTTNG_DOMAIN_KERNEL
:
3215 if (uatomic_read(&kernel_consumerd_state
) != CONSUMER_STARTED
) {
3216 ret
= LTTNG_ERR_NO_KERNCONSUMERD
;
3221 ret
= LTTNG_ERR_UNKNOWN_DOMAIN
;
3227 * Check that the UID or GID match that of the tracing session.
3228 * The root user can interact with all sessions.
3230 if (need_tracing_session
) {
3231 if (!session_access_ok(cmd_ctx
->session
,
3232 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx
->creds
),
3233 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx
->creds
))) {
3234 ret
= LTTNG_ERR_EPERM
;
3240 * Send relayd information to consumer as soon as we have a domain and a
3243 if (cmd_ctx
->session
&& need_domain
) {
3245 * Setup relayd if not done yet. If the relayd information was already
3246 * sent to the consumer, this call will gracefully return.
3248 ret
= cmd_setup_relayd(cmd_ctx
->session
);
3249 if (ret
!= LTTNG_OK
) {
3254 /* Process by command type */
3255 switch (cmd_ctx
->lsm
->cmd_type
) {
3256 case LTTNG_ADD_CONTEXT
:
3259 * An LTTNG_ADD_CONTEXT command might have a supplementary
3260 * payload if the context being added is an application context.
3262 if (cmd_ctx
->lsm
->u
.context
.ctx
.ctx
==
3263 LTTNG_EVENT_CONTEXT_APP_CONTEXT
) {
3264 char *provider_name
= NULL
, *context_name
= NULL
;
3265 size_t provider_name_len
=
3266 cmd_ctx
->lsm
->u
.context
.provider_name_len
;
3267 size_t context_name_len
=
3268 cmd_ctx
->lsm
->u
.context
.context_name_len
;
3270 if (provider_name_len
== 0 || context_name_len
== 0) {
3272 * Application provider and context names MUST
3275 ret
= -LTTNG_ERR_INVALID
;
3279 provider_name
= zmalloc(provider_name_len
+ 1);
3280 if (!provider_name
) {
3281 ret
= -LTTNG_ERR_NOMEM
;
3284 cmd_ctx
->lsm
->u
.context
.ctx
.u
.app_ctx
.provider_name
=
3287 context_name
= zmalloc(context_name_len
+ 1);
3288 if (!context_name
) {
3289 ret
= -LTTNG_ERR_NOMEM
;
3290 goto error_add_context
;
3292 cmd_ctx
->lsm
->u
.context
.ctx
.u
.app_ctx
.ctx_name
=
3295 ret
= lttcomm_recv_unix_sock(sock
, provider_name
,
3298 goto error_add_context
;
3301 ret
= lttcomm_recv_unix_sock(sock
, context_name
,
3304 goto error_add_context
;
3309 * cmd_add_context assumes ownership of the provider and context
3312 ret
= cmd_add_context(cmd_ctx
->session
,
3313 cmd_ctx
->lsm
->domain
.type
,
3314 cmd_ctx
->lsm
->u
.context
.channel_name
,
3315 &cmd_ctx
->lsm
->u
.context
.ctx
,
3316 kernel_poll_pipe
[1]);
3318 cmd_ctx
->lsm
->u
.context
.ctx
.u
.app_ctx
.provider_name
= NULL
;
3319 cmd_ctx
->lsm
->u
.context
.ctx
.u
.app_ctx
.ctx_name
= NULL
;
3321 free(cmd_ctx
->lsm
->u
.context
.ctx
.u
.app_ctx
.provider_name
);
3322 free(cmd_ctx
->lsm
->u
.context
.ctx
.u
.app_ctx
.ctx_name
);
3328 case LTTNG_DISABLE_CHANNEL
:
3330 ret
= cmd_disable_channel(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
3331 cmd_ctx
->lsm
->u
.disable
.channel_name
);
3334 case LTTNG_DISABLE_EVENT
:
3338 * FIXME: handle filter; for now we just receive the filter's
3339 * bytecode along with the filter expression which are sent by
3340 * liblttng-ctl and discard them.
3342 * This fixes an issue where the client may block while sending
3343 * the filter payload and encounter an error because the session
3344 * daemon closes the socket without ever handling this data.
3346 size_t count
= cmd_ctx
->lsm
->u
.disable
.expression_len
+
3347 cmd_ctx
->lsm
->u
.disable
.bytecode_len
;
3350 char data
[LTTNG_FILTER_MAX_LEN
];
3352 DBG("Discarding disable event command payload of size %zu", count
);
3354 ret
= lttcomm_recv_unix_sock(sock
, data
,
3355 count
> sizeof(data
) ? sizeof(data
) : count
);
3360 count
-= (size_t) ret
;
3363 /* FIXME: passing packed structure to non-packed pointer */
3364 ret
= cmd_disable_event(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
3365 cmd_ctx
->lsm
->u
.disable
.channel_name
,
3366 &cmd_ctx
->lsm
->u
.disable
.event
);
3369 case LTTNG_ENABLE_CHANNEL
:
3371 cmd_ctx
->lsm
->u
.channel
.chan
.attr
.extended
.ptr
=
3372 (struct lttng_channel_extended
*) &cmd_ctx
->lsm
->u
.channel
.extended
;
3373 ret
= cmd_enable_channel(cmd_ctx
->session
, &cmd_ctx
->lsm
->domain
,
3374 &cmd_ctx
->lsm
->u
.channel
.chan
,
3375 kernel_poll_pipe
[1]);
3378 case LTTNG_TRACK_PID
:
3380 ret
= cmd_track_pid(cmd_ctx
->session
,
3381 cmd_ctx
->lsm
->domain
.type
,
3382 cmd_ctx
->lsm
->u
.pid_tracker
.pid
);
3385 case LTTNG_UNTRACK_PID
:
3387 ret
= cmd_untrack_pid(cmd_ctx
->session
,
3388 cmd_ctx
->lsm
->domain
.type
,
3389 cmd_ctx
->lsm
->u
.pid_tracker
.pid
);
3392 case LTTNG_ENABLE_EVENT
:
3394 struct lttng_event_exclusion
*exclusion
= NULL
;
3395 struct lttng_filter_bytecode
*bytecode
= NULL
;
3396 char *filter_expression
= NULL
;
3398 /* Handle exclusion events and receive it from the client. */
3399 if (cmd_ctx
->lsm
->u
.enable
.exclusion_count
> 0) {
3400 size_t count
= cmd_ctx
->lsm
->u
.enable
.exclusion_count
;
3402 exclusion
= zmalloc(sizeof(struct lttng_event_exclusion
) +
3403 (count
* LTTNG_SYMBOL_NAME_LEN
));
3405 ret
= LTTNG_ERR_EXCLUSION_NOMEM
;
3409 DBG("Receiving var len exclusion event list from client ...");
3410 exclusion
->count
= count
;
3411 ret
= lttcomm_recv_unix_sock(sock
, exclusion
->names
,
3412 count
* LTTNG_SYMBOL_NAME_LEN
);
3414 DBG("Nothing recv() from client var len data... continuing");
3417 ret
= LTTNG_ERR_EXCLUSION_INVAL
;
3422 /* Get filter expression from client. */
3423 if (cmd_ctx
->lsm
->u
.enable
.expression_len
> 0) {
3424 size_t expression_len
=
3425 cmd_ctx
->lsm
->u
.enable
.expression_len
;
3427 if (expression_len
> LTTNG_FILTER_MAX_LEN
) {
3428 ret
= LTTNG_ERR_FILTER_INVAL
;
3433 filter_expression
= zmalloc(expression_len
);
3434 if (!filter_expression
) {
3436 ret
= LTTNG_ERR_FILTER_NOMEM
;
3440 /* Receive var. len. data */
3441 DBG("Receiving var len filter's expression from client ...");
3442 ret
= lttcomm_recv_unix_sock(sock
, filter_expression
,
3445 DBG("Nothing recv() from client car len data... continuing");
3447 free(filter_expression
);
3449 ret
= LTTNG_ERR_FILTER_INVAL
;
3454 /* Handle filter and get bytecode from client. */
3455 if (cmd_ctx
->lsm
->u
.enable
.bytecode_len
> 0) {
3456 size_t bytecode_len
= cmd_ctx
->lsm
->u
.enable
.bytecode_len
;
3458 if (bytecode_len
> LTTNG_FILTER_MAX_LEN
) {
3459 ret
= LTTNG_ERR_FILTER_INVAL
;
3460 free(filter_expression
);
3465 bytecode
= zmalloc(bytecode_len
);
3467 free(filter_expression
);
3469 ret
= LTTNG_ERR_FILTER_NOMEM
;
3473 /* Receive var. len. data */
3474 DBG("Receiving var len filter's bytecode from client ...");
3475 ret
= lttcomm_recv_unix_sock(sock
, bytecode
, bytecode_len
);
3477 DBG("Nothing recv() from client car len data... continuing");
3479 free(filter_expression
);
3482 ret
= LTTNG_ERR_FILTER_INVAL
;
3486 if ((bytecode
->len
+ sizeof(*bytecode
)) != bytecode_len
) {
3487 free(filter_expression
);
3490 ret
= LTTNG_ERR_FILTER_INVAL
;
3495 ret
= cmd_enable_event(cmd_ctx
->session
, &cmd_ctx
->lsm
->domain
,
3496 cmd_ctx
->lsm
->u
.enable
.channel_name
,
3497 &cmd_ctx
->lsm
->u
.enable
.event
,
3498 filter_expression
, bytecode
, exclusion
,
3499 kernel_poll_pipe
[1]);
3502 case LTTNG_LIST_TRACEPOINTS
:
3504 struct lttng_event
*events
;
3507 session_lock_list();
3508 nb_events
= cmd_list_tracepoints(cmd_ctx
->lsm
->domain
.type
, &events
);
3509 session_unlock_list();
3510 if (nb_events
< 0) {
3511 /* Return value is a negative lttng_error_code. */
3517 * Setup lttng message with payload size set to the event list size in
3518 * bytes and then copy list into the llm payload.
3520 ret
= setup_lttng_msg_no_cmd_header(cmd_ctx
, events
,
3521 sizeof(struct lttng_event
) * nb_events
);
3531 case LTTNG_LIST_TRACEPOINT_FIELDS
:
3533 struct lttng_event_field
*fields
;
3536 session_lock_list();
3537 nb_fields
= cmd_list_tracepoint_fields(cmd_ctx
->lsm
->domain
.type
,
3539 session_unlock_list();
3540 if (nb_fields
< 0) {
3541 /* Return value is a negative lttng_error_code. */
3547 * Setup lttng message with payload size set to the event list size in
3548 * bytes and then copy list into the llm payload.
3550 ret
= setup_lttng_msg_no_cmd_header(cmd_ctx
, fields
,
3551 sizeof(struct lttng_event_field
) * nb_fields
);
3561 case LTTNG_LIST_SYSCALLS
:
3563 struct lttng_event
*events
;
3566 nb_events
= cmd_list_syscalls(&events
);
3567 if (nb_events
< 0) {
3568 /* Return value is a negative lttng_error_code. */
3574 * Setup lttng message with payload size set to the event list size in
3575 * bytes and then copy list into the llm payload.
3577 ret
= setup_lttng_msg_no_cmd_header(cmd_ctx
, events
,
3578 sizeof(struct lttng_event
) * nb_events
);
3588 case LTTNG_LIST_TRACKER_PIDS
:
3590 int32_t *pids
= NULL
;
3593 nr_pids
= cmd_list_tracker_pids(cmd_ctx
->session
,
3594 cmd_ctx
->lsm
->domain
.type
, &pids
);
3596 /* Return value is a negative lttng_error_code. */
3602 * Setup lttng message with payload size set to the event list size in
3603 * bytes and then copy list into the llm payload.
3605 ret
= setup_lttng_msg_no_cmd_header(cmd_ctx
, pids
,
3606 sizeof(int32_t) * nr_pids
);
3616 case LTTNG_SET_CONSUMER_URI
:
3619 struct lttng_uri
*uris
;
3621 nb_uri
= cmd_ctx
->lsm
->u
.uri
.size
;
3622 len
= nb_uri
* sizeof(struct lttng_uri
);
3625 ret
= LTTNG_ERR_INVALID
;
3629 uris
= zmalloc(len
);
3631 ret
= LTTNG_ERR_FATAL
;
3635 /* Receive variable len data */
3636 DBG("Receiving %zu URI(s) from client ...", nb_uri
);
3637 ret
= lttcomm_recv_unix_sock(sock
, uris
, len
);
3639 DBG("No URIs received from client... continuing");
3641 ret
= LTTNG_ERR_SESSION_FAIL
;
3646 ret
= cmd_set_consumer_uri(cmd_ctx
->session
, nb_uri
, uris
);
3648 if (ret
!= LTTNG_OK
) {
3655 case LTTNG_START_TRACE
:
3657 ret
= cmd_start_trace(cmd_ctx
->session
);
3660 case LTTNG_STOP_TRACE
:
3662 ret
= cmd_stop_trace(cmd_ctx
->session
);
3665 case LTTNG_CREATE_SESSION
:
3668 struct lttng_uri
*uris
= NULL
;
3670 nb_uri
= cmd_ctx
->lsm
->u
.uri
.size
;
3671 len
= nb_uri
* sizeof(struct lttng_uri
);
3674 uris
= zmalloc(len
);
3676 ret
= LTTNG_ERR_FATAL
;
3680 /* Receive variable len data */
3681 DBG("Waiting for %zu URIs from client ...", nb_uri
);
3682 ret
= lttcomm_recv_unix_sock(sock
, uris
, len
);
3684 DBG("No URIs received from client... continuing");
3686 ret
= LTTNG_ERR_SESSION_FAIL
;
3691 if (nb_uri
== 1 && uris
[0].dtype
!= LTTNG_DST_PATH
) {
3692 DBG("Creating session with ONE network URI is a bad call");
3693 ret
= LTTNG_ERR_SESSION_FAIL
;
3699 ret
= cmd_create_session_uri(cmd_ctx
->lsm
->session
.name
, uris
, nb_uri
,
3700 &cmd_ctx
->creds
, 0);
3706 case LTTNG_DESTROY_SESSION
:
3708 ret
= cmd_destroy_session(cmd_ctx
->session
, kernel_poll_pipe
[1]);
3710 /* Set session to NULL so we do not unlock it after free. */
3711 cmd_ctx
->session
= NULL
;
3714 case LTTNG_LIST_DOMAINS
:
3717 struct lttng_domain
*domains
= NULL
;
3719 nb_dom
= cmd_list_domains(cmd_ctx
->session
, &domains
);
3721 /* Return value is a negative lttng_error_code. */
3726 ret
= setup_lttng_msg_no_cmd_header(cmd_ctx
, domains
,
3727 nb_dom
* sizeof(struct lttng_domain
));
3737 case LTTNG_LIST_CHANNELS
:
3739 ssize_t payload_size
;
3740 struct lttng_channel
*channels
= NULL
;
3742 payload_size
= cmd_list_channels(cmd_ctx
->lsm
->domain
.type
,
3743 cmd_ctx
->session
, &channels
);
3744 if (payload_size
< 0) {
3745 /* Return value is a negative lttng_error_code. */
3746 ret
= -payload_size
;
3750 ret
= setup_lttng_msg_no_cmd_header(cmd_ctx
, channels
,
3761 case LTTNG_LIST_EVENTS
:
3764 struct lttng_event
*events
= NULL
;
3765 struct lttcomm_event_command_header cmd_header
;
3768 memset(&cmd_header
, 0, sizeof(cmd_header
));
3769 /* Extended infos are included at the end of events */
3770 nb_event
= cmd_list_events(cmd_ctx
->lsm
->domain
.type
,
3771 cmd_ctx
->session
, cmd_ctx
->lsm
->u
.list
.channel_name
,
3772 &events
, &total_size
);
3775 /* Return value is a negative lttng_error_code. */
3780 cmd_header
.nb_events
= nb_event
;
3781 ret
= setup_lttng_msg(cmd_ctx
, events
, total_size
,
3782 &cmd_header
, sizeof(cmd_header
));
3792 case LTTNG_LIST_SESSIONS
:
3794 unsigned int nr_sessions
;
3795 void *sessions_payload
;
3798 session_lock_list();
3799 nr_sessions
= lttng_sessions_count(
3800 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx
->creds
),
3801 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx
->creds
));
3802 payload_len
= sizeof(struct lttng_session
) * nr_sessions
;
3803 sessions_payload
= zmalloc(payload_len
);
3805 if (!sessions_payload
) {
3806 session_unlock_list();
3811 cmd_list_lttng_sessions(sessions_payload
,
3812 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx
->creds
),
3813 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx
->creds
));
3814 session_unlock_list();
3816 ret
= setup_lttng_msg_no_cmd_header(cmd_ctx
, sessions_payload
,
3818 free(sessions_payload
);
3827 case LTTNG_REGISTER_CONSUMER
:
3829 struct consumer_data
*cdata
;
3831 switch (cmd_ctx
->lsm
->domain
.type
) {
3832 case LTTNG_DOMAIN_KERNEL
:
3833 cdata
= &kconsumer_data
;
3836 ret
= LTTNG_ERR_UND
;
3840 ret
= cmd_register_consumer(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
3841 cmd_ctx
->lsm
->u
.reg
.path
, cdata
);
3844 case LTTNG_DATA_PENDING
:
3847 uint8_t pending_ret_byte
;
3849 pending_ret
= cmd_data_pending(cmd_ctx
->session
);
3854 * This function may returns 0 or 1 to indicate whether or not
3855 * there is data pending. In case of error, it should return an
3856 * LTTNG_ERR code. However, some code paths may still return
3857 * a nondescript error code, which we handle by returning an
3860 if (pending_ret
== 0 || pending_ret
== 1) {
3862 * ret will be set to LTTNG_OK at the end of
3865 } else if (pending_ret
< 0) {
3866 ret
= LTTNG_ERR_UNK
;
3873 pending_ret_byte
= (uint8_t) pending_ret
;
3875 /* 1 byte to return whether or not data is pending */
3876 ret
= setup_lttng_msg_no_cmd_header(cmd_ctx
,
3877 &pending_ret_byte
, 1);
3886 case LTTNG_SNAPSHOT_ADD_OUTPUT
:
3888 struct lttcomm_lttng_output_id reply
;
3890 ret
= cmd_snapshot_add_output(cmd_ctx
->session
,
3891 &cmd_ctx
->lsm
->u
.snapshot_output
.output
, &reply
.id
);
3892 if (ret
!= LTTNG_OK
) {
3896 ret
= setup_lttng_msg_no_cmd_header(cmd_ctx
, &reply
,
3902 /* Copy output list into message payload */
3906 case LTTNG_SNAPSHOT_DEL_OUTPUT
:
3908 ret
= cmd_snapshot_del_output(cmd_ctx
->session
,
3909 &cmd_ctx
->lsm
->u
.snapshot_output
.output
);
3912 case LTTNG_SNAPSHOT_LIST_OUTPUT
:
3915 struct lttng_snapshot_output
*outputs
= NULL
;
3917 nb_output
= cmd_snapshot_list_outputs(cmd_ctx
->session
, &outputs
);
3918 if (nb_output
< 0) {
3923 assert((nb_output
> 0 && outputs
) || nb_output
== 0);
3924 ret
= setup_lttng_msg_no_cmd_header(cmd_ctx
, outputs
,
3925 nb_output
* sizeof(struct lttng_snapshot_output
));
3935 case LTTNG_SNAPSHOT_RECORD
:
3937 ret
= cmd_snapshot_record(cmd_ctx
->session
,
3938 &cmd_ctx
->lsm
->u
.snapshot_record
.output
,
3939 cmd_ctx
->lsm
->u
.snapshot_record
.wait
);
3942 case LTTNG_CREATE_SESSION_SNAPSHOT
:
3945 struct lttng_uri
*uris
= NULL
;
3947 nb_uri
= cmd_ctx
->lsm
->u
.uri
.size
;
3948 len
= nb_uri
* sizeof(struct lttng_uri
);
3951 uris
= zmalloc(len
);
3953 ret
= LTTNG_ERR_FATAL
;
3957 /* Receive variable len data */
3958 DBG("Waiting for %zu URIs from client ...", nb_uri
);
3959 ret
= lttcomm_recv_unix_sock(sock
, uris
, len
);
3961 DBG("No URIs received from client... continuing");
3963 ret
= LTTNG_ERR_SESSION_FAIL
;
3968 if (nb_uri
== 1 && uris
[0].dtype
!= LTTNG_DST_PATH
) {
3969 DBG("Creating session with ONE network URI is a bad call");
3970 ret
= LTTNG_ERR_SESSION_FAIL
;
3976 ret
= cmd_create_session_snapshot(cmd_ctx
->lsm
->session
.name
, uris
,
3977 nb_uri
, &cmd_ctx
->creds
);
3981 case LTTNG_CREATE_SESSION_LIVE
:
3984 struct lttng_uri
*uris
= NULL
;
3986 nb_uri
= cmd_ctx
->lsm
->u
.uri
.size
;
3987 len
= nb_uri
* sizeof(struct lttng_uri
);
3990 uris
= zmalloc(len
);
3992 ret
= LTTNG_ERR_FATAL
;
3996 /* Receive variable len data */
3997 DBG("Waiting for %zu URIs from client ...", nb_uri
);
3998 ret
= lttcomm_recv_unix_sock(sock
, uris
, len
);
4000 DBG("No URIs received from client... continuing");
4002 ret
= LTTNG_ERR_SESSION_FAIL
;
4007 if (nb_uri
== 1 && uris
[0].dtype
!= LTTNG_DST_PATH
) {
4008 DBG("Creating session with ONE network URI is a bad call");
4009 ret
= LTTNG_ERR_SESSION_FAIL
;
4015 ret
= cmd_create_session_uri(cmd_ctx
->lsm
->session
.name
, uris
,
4016 nb_uri
, &cmd_ctx
->creds
, cmd_ctx
->lsm
->u
.session_live
.timer_interval
);
4020 case LTTNG_SAVE_SESSION
:
4022 ret
= cmd_save_sessions(&cmd_ctx
->lsm
->u
.save_session
.attr
,
4026 case LTTNG_SET_SESSION_SHM_PATH
:
4028 ret
= cmd_set_session_shm_path(cmd_ctx
->session
,
4029 cmd_ctx
->lsm
->u
.set_shm_path
.shm_path
);
4032 case LTTNG_REGENERATE_METADATA
:
4034 ret
= cmd_regenerate_metadata(cmd_ctx
->session
);
4037 case LTTNG_REGENERATE_STATEDUMP
:
4039 ret
= cmd_regenerate_statedump(cmd_ctx
->session
);
4042 case LTTNG_REGISTER_TRIGGER
:
4044 ret
= cmd_register_trigger(cmd_ctx
, sock
,
4045 notification_thread_handle
);
4048 case LTTNG_UNREGISTER_TRIGGER
:
4050 ret
= cmd_unregister_trigger(cmd_ctx
, sock
,
4051 notification_thread_handle
);
4055 ret
= LTTNG_ERR_UND
;
4060 if (cmd_ctx
->llm
== NULL
) {
4061 DBG("Missing llm structure. Allocating one.");
4062 if (setup_lttng_msg_no_cmd_header(cmd_ctx
, NULL
, 0) < 0) {
4066 /* Set return code */
4067 cmd_ctx
->llm
->ret_code
= ret
;
4069 if (cmd_ctx
->session
) {
4070 session_unlock(cmd_ctx
->session
);
4072 if (need_tracing_session
) {
4073 session_unlock_list();
4076 assert(!rcu_read_ongoing());
4081 * Thread managing health check socket.
4083 static void *thread_manage_health(void *data
)
4085 int sock
= -1, new_sock
= -1, ret
, i
, pollfd
, err
= -1;
4086 uint32_t revents
, nb_fd
;
4087 struct lttng_poll_event events
;
4088 struct health_comm_msg msg
;
4089 struct health_comm_reply reply
;
4091 DBG("[thread] Manage health check started");
4093 rcu_register_thread();
4095 /* We might hit an error path before this is created. */
4096 lttng_poll_init(&events
);
4098 /* Create unix socket */
4099 sock
= lttcomm_create_unix_sock(config
.health_unix_sock_path
.value
);
4101 ERR("Unable to create health check Unix socket");
4106 /* lttng health client socket path permissions */
4107 ret
= chown(config
.health_unix_sock_path
.value
, 0,
4108 utils_get_group_id(config
.tracing_group_name
.value
));
4110 ERR("Unable to set group on %s", config
.health_unix_sock_path
.value
);
4115 ret
= chmod(config
.health_unix_sock_path
.value
,
4116 S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IWGRP
);
4118 ERR("Unable to set permissions on %s", config
.health_unix_sock_path
.value
);
4125 * Set the CLOEXEC flag. Return code is useless because either way, the
4128 (void) utils_set_fd_cloexec(sock
);
4130 ret
= lttcomm_listen_unix_sock(sock
);
4136 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
4137 * more will be added to this poll set.
4139 ret
= sessiond_set_thread_pollset(&events
, 2);
4144 /* Add the application registration socket */
4145 ret
= lttng_poll_add(&events
, sock
, LPOLLIN
| LPOLLPRI
);
4150 sessiond_notify_ready();
4153 DBG("Health check ready");
4155 /* Inifinite blocking call, waiting for transmission */
4157 ret
= lttng_poll_wait(&events
, -1);
4160 * Restart interrupted system call.
4162 if (errno
== EINTR
) {
4170 for (i
= 0; i
< nb_fd
; i
++) {
4171 /* Fetch once the poll data */
4172 revents
= LTTNG_POLL_GETEV(&events
, i
);
4173 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
4176 /* No activity for this FD (poll implementation). */
4180 /* Thread quit pipe has been closed. Killing thread. */
4181 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
4187 /* Event on the registration socket */
4188 if (pollfd
== sock
) {
4189 if (revents
& LPOLLIN
) {
4191 } else if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
4192 ERR("Health socket poll error");
4195 ERR("Unexpected poll events %u for sock %d", revents
, pollfd
);
4201 new_sock
= lttcomm_accept_unix_sock(sock
);
4207 * Set the CLOEXEC flag. Return code is useless because either way, the
4210 (void) utils_set_fd_cloexec(new_sock
);
4212 DBG("Receiving data from client for health...");
4213 ret
= lttcomm_recv_unix_sock(new_sock
, (void *)&msg
, sizeof(msg
));
4215 DBG("Nothing recv() from client... continuing");
4216 ret
= close(new_sock
);
4223 rcu_thread_online();
4225 memset(&reply
, 0, sizeof(reply
));
4226 for (i
= 0; i
< NR_HEALTH_SESSIOND_TYPES
; i
++) {
4228 * health_check_state returns 0 if health is
4231 if (!health_check_state(health_sessiond
, i
)) {
4232 reply
.ret_code
|= 1ULL << i
;
4236 DBG2("Health check return value %" PRIx64
, reply
.ret_code
);
4238 ret
= send_unix_sock(new_sock
, (void *) &reply
, sizeof(reply
));
4240 ERR("Failed to send health data back to client");
4243 /* End of transmission */
4244 ret
= close(new_sock
);
4253 ERR("Health error occurred in %s", __func__
);
4255 DBG("Health check thread dying");
4256 unlink(config
.health_unix_sock_path
.value
);
4264 lttng_poll_clean(&events
);
4266 rcu_unregister_thread();
4271 * This thread manage all clients request using the unix client socket for
4274 static void *thread_manage_clients(void *data
)
4276 int sock
= -1, ret
, i
, pollfd
, err
= -1;
4278 uint32_t revents
, nb_fd
;
4279 struct command_ctx
*cmd_ctx
= NULL
;
4280 struct lttng_poll_event events
;
4282 DBG("[thread] Manage client started");
4284 rcu_register_thread();
4286 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_CMD
);
4288 health_code_update();
4290 ret
= lttcomm_listen_unix_sock(client_sock
);
4296 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
4297 * more will be added to this poll set.
4299 ret
= sessiond_set_thread_pollset(&events
, 2);
4301 goto error_create_poll
;
4304 /* Add the application registration socket */
4305 ret
= lttng_poll_add(&events
, client_sock
, LPOLLIN
| LPOLLPRI
);
4310 sessiond_notify_ready();
4311 ret
= sem_post(&load_info
->message_thread_ready
);
4313 PERROR("sem_post message_thread_ready");
4317 /* This testpoint is after we signal readiness to the parent. */
4318 if (testpoint(sessiond_thread_manage_clients
)) {
4322 if (testpoint(sessiond_thread_manage_clients_before_loop
)) {
4326 health_code_update();
4329 DBG("Accepting client command ...");
4331 /* Inifinite blocking call, waiting for transmission */
4333 health_poll_entry();
4334 ret
= lttng_poll_wait(&events
, -1);
4338 * Restart interrupted system call.
4340 if (errno
== EINTR
) {
4348 for (i
= 0; i
< nb_fd
; i
++) {
4349 /* Fetch once the poll data */
4350 revents
= LTTNG_POLL_GETEV(&events
, i
);
4351 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
4353 health_code_update();
4356 /* No activity for this FD (poll implementation). */
4360 /* Thread quit pipe has been closed. Killing thread. */
4361 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
4367 /* Event on the registration socket */
4368 if (pollfd
== client_sock
) {
4369 if (revents
& LPOLLIN
) {
4371 } else if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
4372 ERR("Client socket poll error");
4375 ERR("Unexpected poll events %u for sock %d", revents
, pollfd
);
4381 DBG("Wait for client response");
4383 health_code_update();
4385 sock
= lttcomm_accept_unix_sock(client_sock
);
4391 * Set the CLOEXEC flag. Return code is useless because either way, the
4394 (void) utils_set_fd_cloexec(sock
);
4396 /* Set socket option for credentials retrieval */
4397 ret
= lttcomm_setsockopt_creds_unix_sock(sock
);
4402 /* Allocate context command to process the client request */
4403 cmd_ctx
= zmalloc(sizeof(struct command_ctx
));
4404 if (cmd_ctx
== NULL
) {
4405 PERROR("zmalloc cmd_ctx");
4409 /* Allocate data buffer for reception */
4410 cmd_ctx
->lsm
= zmalloc(sizeof(struct lttcomm_session_msg
));
4411 if (cmd_ctx
->lsm
== NULL
) {
4412 PERROR("zmalloc cmd_ctx->lsm");
4416 cmd_ctx
->llm
= NULL
;
4417 cmd_ctx
->session
= NULL
;
4419 health_code_update();
4422 * Data is received from the lttng client. The struct
4423 * lttcomm_session_msg (lsm) contains the command and data request of
4426 DBG("Receiving data from client ...");
4427 ret
= lttcomm_recv_creds_unix_sock(sock
, cmd_ctx
->lsm
,
4428 sizeof(struct lttcomm_session_msg
), &cmd_ctx
->creds
);
4430 DBG("Nothing recv() from client... continuing");
4436 clean_command_ctx(&cmd_ctx
);
4440 health_code_update();
4442 // TODO: Validate cmd_ctx including sanity check for
4443 // security purpose.
4445 rcu_thread_online();
4447 * This function dispatch the work to the kernel or userspace tracer
4448 * libs and fill the lttcomm_lttng_msg data structure of all the needed
4449 * informations for the client. The command context struct contains
4450 * everything this function may needs.
4452 ret
= process_client_msg(cmd_ctx
, sock
, &sock_error
);
4453 rcu_thread_offline();
4461 * TODO: Inform client somehow of the fatal error. At
4462 * this point, ret < 0 means that a zmalloc failed
4463 * (ENOMEM). Error detected but still accept
4464 * command, unless a socket error has been
4467 clean_command_ctx(&cmd_ctx
);
4471 health_code_update();
4473 DBG("Sending response (size: %d, retcode: %s (%d))",
4474 cmd_ctx
->lttng_msg_size
,
4475 lttng_strerror(-cmd_ctx
->llm
->ret_code
),
4476 cmd_ctx
->llm
->ret_code
);
4477 ret
= send_unix_sock(sock
, cmd_ctx
->llm
, cmd_ctx
->lttng_msg_size
);
4479 ERR("Failed to send data back to client");
4482 /* End of transmission */
4489 clean_command_ctx(&cmd_ctx
);
4491 health_code_update();
4503 lttng_poll_clean(&events
);
4504 clean_command_ctx(&cmd_ctx
);
4508 unlink(config
.client_unix_sock_path
.value
);
4509 if (client_sock
>= 0) {
4510 ret
= close(client_sock
);
4518 ERR("Health error occurred in %s", __func__
);
4521 health_unregister(health_sessiond
);
4523 DBG("Client thread dying");
4525 rcu_unregister_thread();
4528 * Since we are creating the consumer threads, we own them, so we need
4529 * to join them before our thread exits.
4531 ret
= join_consumer_thread(&kconsumer_data
);
4534 PERROR("join_consumer");
4537 ret
= join_consumer_thread(&ustconsumer32_data
);
4540 PERROR("join_consumer ust32");
4543 ret
= join_consumer_thread(&ustconsumer64_data
);
4546 PERROR("join_consumer ust64");
4551 static int string_match(const char *str1
, const char *str2
)
4553 return (str1
&& str2
) && !strcmp(str1
, str2
);
4557 * Take an option from the getopt output and set it in the right variable to be
4560 * Return 0 on success else a negative value.
4562 static int set_option(int opt
, const char *arg
, const char *optname
)
4566 if (string_match(optname
, "client-sock") || opt
== 'c') {
4567 if (!arg
|| *arg
== '\0') {
4571 if (lttng_is_setuid_setgid()) {
4572 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4573 "-c, --client-sock");
4575 config_string_set(&config
.client_unix_sock_path
,
4577 if (!config
.client_unix_sock_path
.value
) {
4582 } else if (string_match(optname
, "apps-sock") || opt
== 'a') {
4583 if (!arg
|| *arg
== '\0') {
4587 if (lttng_is_setuid_setgid()) {
4588 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4591 config_string_set(&config
.apps_unix_sock_path
,
4593 if (!config
.apps_unix_sock_path
.value
) {
4598 } else if (string_match(optname
, "daemonize") || opt
== 'd') {
4599 config
.daemonize
= true;
4600 } else if (string_match(optname
, "background") || opt
== 'b') {
4601 config
.background
= true;
4602 } else if (string_match(optname
, "group") || opt
== 'g') {
4603 if (!arg
|| *arg
== '\0') {
4607 if (lttng_is_setuid_setgid()) {
4608 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4611 config_string_set(&config
.tracing_group_name
,
4613 if (!config
.tracing_group_name
.value
) {
4618 } else if (string_match(optname
, "help") || opt
== 'h') {
4619 ret
= utils_show_help(8, "lttng-sessiond", help_msg
);
4621 ERR("Cannot show --help for `lttng-sessiond`");
4624 exit(ret
? EXIT_FAILURE
: EXIT_SUCCESS
);
4625 } else if (string_match(optname
, "version") || opt
== 'V') {
4626 fprintf(stdout
, "%s\n", VERSION
);
4628 } else if (string_match(optname
, "sig-parent") || opt
== 'S') {
4629 config
.sig_parent
= true;
4630 } else if (string_match(optname
, "kconsumerd-err-sock")) {
4631 if (!arg
|| *arg
== '\0') {
4635 if (lttng_is_setuid_setgid()) {
4636 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4637 "--kconsumerd-err-sock");
4639 config_string_set(&config
.kconsumerd_err_unix_sock_path
,
4641 if (!config
.kconsumerd_err_unix_sock_path
.value
) {
4646 } else if (string_match(optname
, "kconsumerd-cmd-sock")) {
4647 if (!arg
|| *arg
== '\0') {
4651 if (lttng_is_setuid_setgid()) {
4652 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4653 "--kconsumerd-cmd-sock");
4655 config_string_set(&config
.kconsumerd_cmd_unix_sock_path
,
4657 if (!config
.kconsumerd_cmd_unix_sock_path
.value
) {
4662 } else if (string_match(optname
, "ustconsumerd64-err-sock")) {
4663 if (!arg
|| *arg
== '\0') {
4667 if (lttng_is_setuid_setgid()) {
4668 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4669 "--ustconsumerd64-err-sock");
4671 config_string_set(&config
.consumerd64_err_unix_sock_path
,
4673 if (!config
.consumerd64_err_unix_sock_path
.value
) {
4678 } else if (string_match(optname
, "ustconsumerd64-cmd-sock")) {
4679 if (!arg
|| *arg
== '\0') {
4683 if (lttng_is_setuid_setgid()) {
4684 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4685 "--ustconsumerd64-cmd-sock");
4687 config_string_set(&config
.consumerd64_cmd_unix_sock_path
,
4689 if (!config
.consumerd64_cmd_unix_sock_path
.value
) {
4694 } else if (string_match(optname
, "ustconsumerd32-err-sock")) {
4695 if (!arg
|| *arg
== '\0') {
4699 if (lttng_is_setuid_setgid()) {
4700 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4701 "--ustconsumerd32-err-sock");
4703 config_string_set(&config
.consumerd32_err_unix_sock_path
,
4705 if (!config
.consumerd32_err_unix_sock_path
.value
) {
4710 } else if (string_match(optname
, "ustconsumerd32-cmd-sock")) {
4711 if (!arg
|| *arg
== '\0') {
4715 if (lttng_is_setuid_setgid()) {
4716 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4717 "--ustconsumerd32-cmd-sock");
4719 config_string_set(&config
.consumerd32_cmd_unix_sock_path
,
4721 if (!config
.consumerd32_cmd_unix_sock_path
.value
) {
4726 } else if (string_match(optname
, "no-kernel")) {
4727 config
.no_kernel
= true;
4728 } else if (string_match(optname
, "quiet") || opt
== 'q') {
4729 lttng_opt_quiet
= true;
4730 } else if (string_match(optname
, "verbose") || opt
== 'v') {
4731 /* Verbose level can increase using multiple -v */
4733 /* Value obtained from config file */
4734 config
.verbose
= config_parse_value(arg
);
4736 /* -v used on command line */
4739 /* Clamp value to [0, 3] */
4740 config
.verbose
= config
.verbose
< 0 ? 0 :
4741 (config
.verbose
<= 3 ? config
.verbose
: 3);
4742 } else if (string_match(optname
, "verbose-consumer")) {
4744 config
.verbose_consumer
= config_parse_value(arg
);
4746 config
.verbose_consumer
++;
4748 } else if (string_match(optname
, "consumerd32-path")) {
4749 if (!arg
|| *arg
== '\0') {
4753 if (lttng_is_setuid_setgid()) {
4754 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4755 "--consumerd32-path");
4757 config_string_set(&config
.consumerd32_bin_path
,
4759 if (!config
.consumerd32_bin_path
.value
) {
4764 } else if (string_match(optname
, "consumerd32-libdir")) {
4765 if (!arg
|| *arg
== '\0') {
4769 if (lttng_is_setuid_setgid()) {
4770 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4771 "--consumerd32-libdir");
4773 config_string_set(&config
.consumerd32_lib_dir
,
4775 if (!config
.consumerd32_lib_dir
.value
) {
4780 } else if (string_match(optname
, "consumerd64-path")) {
4781 if (!arg
|| *arg
== '\0') {
4785 if (lttng_is_setuid_setgid()) {
4786 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4787 "--consumerd64-path");
4789 config_string_set(&config
.consumerd64_bin_path
,
4791 if (!config
.consumerd64_bin_path
.value
) {
4796 } else if (string_match(optname
, "consumerd64-libdir")) {
4797 if (!arg
|| *arg
== '\0') {
4801 if (lttng_is_setuid_setgid()) {
4802 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4803 "--consumerd64-libdir");
4805 config_string_set(&config
.consumerd64_lib_dir
,
4807 if (!config
.consumerd64_lib_dir
.value
) {
4812 } else if (string_match(optname
, "pidfile") || opt
== 'p') {
4813 if (!arg
|| *arg
== '\0') {
4817 if (lttng_is_setuid_setgid()) {
4818 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4821 config_string_set(&config
.pid_file_path
, strdup(arg
));
4822 if (!config
.pid_file_path
.value
) {
4827 } else if (string_match(optname
, "agent-tcp-port")) {
4828 if (!arg
|| *arg
== '\0') {
4832 if (lttng_is_setuid_setgid()) {
4833 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4834 "--agent-tcp-port");
4839 v
= strtoul(arg
, NULL
, 0);
4840 if (errno
!= 0 || !isdigit(arg
[0])) {
4841 ERR("Wrong value in --agent-tcp-port parameter: %s", arg
);
4844 if (v
== 0 || v
>= 65535) {
4845 ERR("Port overflow in --agent-tcp-port parameter: %s", arg
);
4848 config
.agent_tcp_port
= (uint32_t) v
;
4849 DBG3("Agent TCP port set to non default: %u", config
.agent_tcp_port
);
4851 } else if (string_match(optname
, "load") || opt
== 'l') {
4852 if (!arg
|| *arg
== '\0') {
4856 if (lttng_is_setuid_setgid()) {
4857 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4860 config_string_set(&config
.load_session_path
, strdup(arg
));
4861 if (!config
.load_session_path
.value
) {
4866 } else if (string_match(optname
, "kmod-probes")) {
4867 if (!arg
|| *arg
== '\0') {
4871 if (lttng_is_setuid_setgid()) {
4872 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4875 config_string_set(&config
.kmod_probes_list
, strdup(arg
));
4876 if (!config
.kmod_probes_list
.value
) {
4881 } else if (string_match(optname
, "extra-kmod-probes")) {
4882 if (!arg
|| *arg
== '\0') {
4886 if (lttng_is_setuid_setgid()) {
4887 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4888 "--extra-kmod-probes");
4890 config_string_set(&config
.kmod_extra_probes_list
,
4892 if (!config
.kmod_extra_probes_list
.value
) {
4897 } else if (string_match(optname
, "config") || opt
== 'f') {
4898 /* This is handled in set_options() thus silent skip. */
4901 /* Unknown option or other error.
4902 * Error is printed by getopt, just return */
4907 if (ret
== -EINVAL
) {
4908 const char *opt_name
= "unknown";
4911 for (i
= 0; i
< sizeof(long_options
) / sizeof(struct option
);
4913 if (opt
== long_options
[i
].val
) {
4914 opt_name
= long_options
[i
].name
;
4919 WARN("Invalid argument provided for option \"%s\", using default value.",
4927 * config_entry_handler_cb used to handle options read from a config file.
4928 * See config_entry_handler_cb comment in common/config/session-config.h for the
4929 * return value conventions.
4931 static int config_entry_handler(const struct config_entry
*entry
, void *unused
)
4935 if (!entry
|| !entry
->name
|| !entry
->value
) {
4940 /* Check if the option is to be ignored */
4941 for (i
= 0; i
< sizeof(config_ignore_options
) / sizeof(char *); i
++) {
4942 if (!strcmp(entry
->name
, config_ignore_options
[i
])) {
4947 for (i
= 0; i
< (sizeof(long_options
) / sizeof(struct option
)) - 1;
4950 /* Ignore if not fully matched. */
4951 if (strcmp(entry
->name
, long_options
[i
].name
)) {
4956 * If the option takes no argument on the command line, we have to
4957 * check if the value is "true". We support non-zero numeric values,
4960 if (!long_options
[i
].has_arg
) {
4961 ret
= config_parse_value(entry
->value
);
4964 WARN("Invalid configuration value \"%s\" for option %s",
4965 entry
->value
, entry
->name
);
4967 /* False, skip boolean config option. */
4972 ret
= set_option(long_options
[i
].val
, entry
->value
, entry
->name
);
4976 WARN("Unrecognized option \"%s\" in daemon configuration file.", entry
->name
);
4983 * daemon configuration loading and argument parsing
4985 static int set_options(int argc
, char **argv
)
4987 int ret
= 0, c
= 0, option_index
= 0;
4988 int orig_optopt
= optopt
, orig_optind
= optind
;
4990 const char *config_path
= NULL
;
4992 optstring
= utils_generate_optstring(long_options
,
4993 sizeof(long_options
) / sizeof(struct option
));
4999 /* Check for the --config option */
5000 while ((c
= getopt_long(argc
, argv
, optstring
, long_options
,
5001 &option_index
)) != -1) {
5005 } else if (c
!= 'f') {
5006 /* if not equal to --config option. */
5010 if (lttng_is_setuid_setgid()) {
5011 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
5014 config_path
= utils_expand_path(optarg
);
5016 ERR("Failed to resolve path: %s", optarg
);
5021 ret
= config_get_section_entries(config_path
, config_section_name
,
5022 config_entry_handler
, NULL
);
5025 ERR("Invalid configuration option at line %i", ret
);
5031 /* Reset getopt's global state */
5032 optopt
= orig_optopt
;
5033 optind
= orig_optind
;
5037 * getopt_long() will not set option_index if it encounters a
5040 c
= getopt_long(argc
, argv
, optstring
, long_options
,
5047 * Pass NULL as the long option name if popt left the index
5050 ret
= set_option(c
, optarg
,
5051 option_index
< 0 ? NULL
:
5052 long_options
[option_index
].name
);
5064 * Creates the two needed socket by the daemon.
5065 * apps_sock - The communication socket for all UST apps.
5066 * client_sock - The communication of the cli tool (lttng).
5068 static int init_daemon_socket(void)
5073 old_umask
= umask(0);
5075 /* Create client tool unix socket */
5076 client_sock
= lttcomm_create_unix_sock(config
.client_unix_sock_path
.value
);
5077 if (client_sock
< 0) {
5078 ERR("Create unix sock failed: %s", config
.client_unix_sock_path
.value
);
5083 /* Set the cloexec flag */
5084 ret
= utils_set_fd_cloexec(client_sock
);
5086 ERR("Unable to set CLOEXEC flag to the client Unix socket (fd: %d). "
5087 "Continuing but note that the consumer daemon will have a "
5088 "reference to this socket on exec()", client_sock
);
5091 /* File permission MUST be 660 */
5092 ret
= chmod(config
.client_unix_sock_path
.value
, S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IWGRP
);
5094 ERR("Set file permissions failed: %s", config
.client_unix_sock_path
.value
);
5099 /* Create the application unix socket */
5100 apps_sock
= lttcomm_create_unix_sock(config
.apps_unix_sock_path
.value
);
5101 if (apps_sock
< 0) {
5102 ERR("Create unix sock failed: %s", config
.apps_unix_sock_path
.value
);
5107 /* Set the cloexec flag */
5108 ret
= utils_set_fd_cloexec(apps_sock
);
5110 ERR("Unable to set CLOEXEC flag to the app Unix socket (fd: %d). "
5111 "Continuing but note that the consumer daemon will have a "
5112 "reference to this socket on exec()", apps_sock
);
5115 /* File permission MUST be 666 */
5116 ret
= chmod(config
.apps_unix_sock_path
.value
,
5117 S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IWGRP
| S_IROTH
| S_IWOTH
);
5119 ERR("Set file permissions failed: %s", config
.apps_unix_sock_path
.value
);
5124 DBG3("Session daemon client socket %d and application socket %d created",
5125 client_sock
, apps_sock
);
5133 * Check if the global socket is available, and if a daemon is answering at the
5134 * other side. If yes, error is returned.
5136 static int check_existing_daemon(void)
5138 /* Is there anybody out there ? */
5139 if (lttng_session_daemon_alive()) {
5147 * Set the tracing group gid onto the client socket.
5149 * Race window between mkdir and chown is OK because we are going from more
5150 * permissive (root.root) to less permissive (root.tracing).
5152 static int set_permissions(char *rundir
)
5157 gid
= utils_get_group_id(config
.tracing_group_name
.value
);
5159 /* Set lttng run dir */
5160 ret
= chown(rundir
, 0, gid
);
5162 ERR("Unable to set group on %s", rundir
);
5167 * Ensure all applications and tracing group can search the run
5168 * dir. Allow everyone to read the directory, since it does not
5169 * buy us anything to hide its content.
5171 ret
= chmod(rundir
, S_IRWXU
| S_IRGRP
| S_IXGRP
| S_IROTH
| S_IXOTH
);
5173 ERR("Unable to set permissions on %s", rundir
);
5177 /* lttng client socket path */
5178 ret
= chown(config
.client_unix_sock_path
.value
, 0, gid
);
5180 ERR("Unable to set group on %s", config
.client_unix_sock_path
.value
);
5184 /* kconsumer error socket path */
5185 ret
= chown(kconsumer_data
.err_unix_sock_path
, 0, 0);
5187 ERR("Unable to set group on %s", kconsumer_data
.err_unix_sock_path
);
5191 /* 64-bit ustconsumer error socket path */
5192 ret
= chown(ustconsumer64_data
.err_unix_sock_path
, 0, 0);
5194 ERR("Unable to set group on %s", ustconsumer64_data
.err_unix_sock_path
);
5198 /* 32-bit ustconsumer compat32 error socket path */
5199 ret
= chown(ustconsumer32_data
.err_unix_sock_path
, 0, 0);
5201 ERR("Unable to set group on %s", ustconsumer32_data
.err_unix_sock_path
);
5205 DBG("All permissions are set");
5211 * Create the lttng run directory needed for all global sockets and pipe.
5213 static int create_lttng_rundir(void)
5217 DBG3("Creating LTTng run directory: %s", config
.rundir
.value
);
5219 ret
= mkdir(config
.rundir
.value
, S_IRWXU
);
5221 if (errno
!= EEXIST
) {
5222 ERR("Unable to create %s", config
.rundir
.value
);
5234 * Setup sockets and directory needed by the consumerds' communication with the
5237 static int set_consumer_sockets(struct consumer_data
*consumer_data
)
5242 switch (consumer_data
->type
) {
5243 case LTTNG_CONSUMER_KERNEL
:
5244 path
= config
.kconsumerd_path
.value
;
5246 case LTTNG_CONSUMER64_UST
:
5247 path
= config
.consumerd64_path
.value
;
5249 case LTTNG_CONSUMER32_UST
:
5250 path
= config
.consumerd32_path
.value
;
5253 ERR("Consumer type unknown");
5259 DBG2("Creating consumer directory: %s", path
);
5261 ret
= mkdir(path
, S_IRWXU
| S_IRGRP
| S_IXGRP
);
5262 if (ret
< 0 && errno
!= EEXIST
) {
5264 ERR("Failed to create %s", path
);
5268 ret
= chown(path
, 0, utils_get_group_id(config
.tracing_group_name
.value
));
5270 ERR("Unable to set group on %s", path
);
5276 /* Create the consumerd error unix socket */
5277 consumer_data
->err_sock
=
5278 lttcomm_create_unix_sock(consumer_data
->err_unix_sock_path
);
5279 if (consumer_data
->err_sock
< 0) {
5280 ERR("Create unix sock failed: %s", consumer_data
->err_unix_sock_path
);
5286 * Set the CLOEXEC flag. Return code is useless because either way, the
5289 ret
= utils_set_fd_cloexec(consumer_data
->err_sock
);
5291 PERROR("utils_set_fd_cloexec");
5292 /* continue anyway */
5295 /* File permission MUST be 660 */
5296 ret
= chmod(consumer_data
->err_unix_sock_path
,
5297 S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IWGRP
);
5299 ERR("Set file permissions failed: %s", consumer_data
->err_unix_sock_path
);
5309 * Signal handler for the daemon
5311 * Simply stop all worker threads, leaving main() return gracefully after
5312 * joining all threads and calling cleanup().
5314 static void sighandler(int sig
)
5318 DBG("SIGINT caught");
5322 DBG("SIGTERM caught");
5326 CMM_STORE_SHARED(recv_child_signal
, 1);
5334 * Setup signal handler for :
5335 * SIGINT, SIGTERM, SIGPIPE
5337 static int set_signal_handler(void)
5340 struct sigaction sa
;
5343 if ((ret
= sigemptyset(&sigset
)) < 0) {
5344 PERROR("sigemptyset");
5348 sa
.sa_mask
= sigset
;
5351 sa
.sa_handler
= sighandler
;
5352 if ((ret
= sigaction(SIGTERM
, &sa
, NULL
)) < 0) {
5353 PERROR("sigaction");
5357 if ((ret
= sigaction(SIGINT
, &sa
, NULL
)) < 0) {
5358 PERROR("sigaction");
5362 if ((ret
= sigaction(SIGUSR1
, &sa
, NULL
)) < 0) {
5363 PERROR("sigaction");
5367 sa
.sa_handler
= SIG_IGN
;
5368 if ((ret
= sigaction(SIGPIPE
, &sa
, NULL
)) < 0) {
5369 PERROR("sigaction");
5373 DBG("Signal handler set for SIGTERM, SIGUSR1, SIGPIPE and SIGINT");
5379 * Set open files limit to unlimited. This daemon can open a large number of
5380 * file descriptors in order to consume multiple kernel traces.
5382 static void set_ulimit(void)
5387 /* The kernel does not allow an infinite limit for open files */
5388 lim
.rlim_cur
= 65535;
5389 lim
.rlim_max
= 65535;
5391 ret
= setrlimit(RLIMIT_NOFILE
, &lim
);
5393 PERROR("failed to set open files limit");
5397 static int write_pidfile(void)
5399 return utils_create_pid_file(getpid(), config
.pid_file_path
.value
);
5403 * Create lockfile using the rundir and return its fd.
5405 static int create_lockfile(void)
5407 return utils_create_lock_file(config
.lock_file_path
.value
);
5411 * Write agent TCP port using the rundir.
5413 static int write_agent_port(void)
5415 return utils_create_pid_file(config
.agent_tcp_port
,
5416 config
.agent_port_file_path
.value
);
5419 static int set_clock_plugin_env(void)
5422 char *env_value
= NULL
;
5424 if (!config
.lttng_ust_clock_plugin
.value
) {
5428 ret
= asprintf(&env_value
, "LTTNG_UST_CLOCK_PLUGIN=%s",
5429 config
.lttng_ust_clock_plugin
.value
);
5435 ret
= putenv(env_value
);
5438 PERROR("putenv of LTTNG_UST_CLOCK_PLUGIN");
5442 DBG("Updated LTTNG_UST_CLOCK_PLUGIN environment variable to \"%s\"",
5443 config
.lttng_ust_clock_plugin
.value
);
5451 int main(int argc
, char **argv
)
5453 int ret
= 0, retval
= 0;
5455 const char *env_app_timeout
;
5456 struct lttng_pipe
*ust32_channel_monitor_pipe
= NULL
,
5457 *ust64_channel_monitor_pipe
= NULL
,
5458 *kernel_channel_monitor_pipe
= NULL
;
5459 bool notification_thread_running
= false;
5461 init_kernel_workarounds();
5463 rcu_register_thread();
5465 if (set_signal_handler()) {
5467 goto exit_set_signal_handler
;
5470 page_size
= sysconf(_SC_PAGESIZE
);
5471 if (page_size
< 0) {
5472 PERROR("sysconf _SC_PAGESIZE");
5473 page_size
= LONG_MAX
;
5474 WARN("Fallback page size to %ld", page_size
);
5477 ret
= sessiond_config_init(&config
);
5480 goto exit_set_signal_handler
;
5484 * Parse arguments and load the daemon configuration file.
5486 * We have an exit_options exit path to free memory reserved by
5487 * set_options. This is needed because the rest of sessiond_cleanup()
5488 * depends on ht_cleanup_thread, which depends on lttng_daemonize, which
5489 * depends on set_options.
5492 if (set_options(argc
, argv
)) {
5497 /* Init config from environment variables. */
5498 sessiond_config_apply_env_config(&config
);
5501 * Resolve all paths received as arguments, configuration option, or
5502 * through environment variable as absolute paths. This is necessary
5503 * since daemonizing causes the sessiond's current working directory
5506 ret
= sessiond_config_resolve_paths(&config
);
5512 lttng_opt_verbose
= config
.verbose
;
5513 lttng_opt_quiet
= config
.quiet
;
5514 kconsumer_data
.err_unix_sock_path
=
5515 config
.kconsumerd_err_unix_sock_path
.value
;
5516 kconsumer_data
.cmd_unix_sock_path
=
5517 config
.kconsumerd_cmd_unix_sock_path
.value
;
5518 ustconsumer32_data
.err_unix_sock_path
=
5519 config
.consumerd32_err_unix_sock_path
.value
;
5520 ustconsumer32_data
.cmd_unix_sock_path
=
5521 config
.consumerd32_cmd_unix_sock_path
.value
;
5522 ustconsumer64_data
.err_unix_sock_path
=
5523 config
.consumerd64_err_unix_sock_path
.value
;
5524 ustconsumer64_data
.cmd_unix_sock_path
=
5525 config
.consumerd64_cmd_unix_sock_path
.value
;
5526 set_clock_plugin_env();
5528 sessiond_config_log(&config
);
5531 if (config
.daemonize
|| config
.background
) {
5534 ret
= lttng_daemonize(&child_ppid
, &recv_child_signal
,
5535 !config
.background
);
5542 * We are in the child. Make sure all other file descriptors are
5543 * closed, in case we are called with more opened file
5544 * descriptors than the standard ones.
5546 for (i
= 3; i
< sysconf(_SC_OPEN_MAX
); i
++) {
5551 if (run_as_create_worker(argv
[0]) < 0) {
5552 goto exit_create_run_as_worker_cleanup
;
5556 * Starting from here, we can create threads. This needs to be after
5557 * lttng_daemonize due to RCU.
5561 * Initialize the health check subsystem. This call should set the
5562 * appropriate time values.
5564 health_sessiond
= health_app_create(NR_HEALTH_SESSIOND_TYPES
);
5565 if (!health_sessiond
) {
5566 PERROR("health_app_create error");
5568 goto exit_health_sessiond_cleanup
;
5571 /* Create thread to clean up RCU hash tables */
5572 if (init_ht_cleanup_thread(&ht_cleanup_thread
)) {
5574 goto exit_ht_cleanup
;
5577 /* Create thread quit pipe */
5578 if (init_thread_quit_pipe()) {
5580 goto exit_init_data
;
5583 /* Check if daemon is UID = 0 */
5584 is_root
= !getuid();
5586 if (create_lttng_rundir()) {
5588 goto exit_init_data
;
5592 /* Create global run dir with root access */
5594 kernel_channel_monitor_pipe
= lttng_pipe_open(0);
5595 if (!kernel_channel_monitor_pipe
) {
5596 ERR("Failed to create kernel consumer channel monitor pipe");
5598 goto exit_init_data
;
5600 kconsumer_data
.channel_monitor_pipe
=
5601 lttng_pipe_release_writefd(
5602 kernel_channel_monitor_pipe
);
5603 if (kconsumer_data
.channel_monitor_pipe
< 0) {
5605 goto exit_init_data
;
5609 lockfile_fd
= create_lockfile();
5610 if (lockfile_fd
< 0) {
5612 goto exit_init_data
;
5615 /* Set consumer initial state */
5616 kernel_consumerd_state
= CONSUMER_STOPPED
;
5617 ust_consumerd_state
= CONSUMER_STOPPED
;
5619 ust32_channel_monitor_pipe
= lttng_pipe_open(0);
5620 if (!ust32_channel_monitor_pipe
) {
5621 ERR("Failed to create 32-bit user space consumer channel monitor pipe");
5623 goto exit_init_data
;
5625 ustconsumer32_data
.channel_monitor_pipe
= lttng_pipe_release_writefd(
5626 ust32_channel_monitor_pipe
);
5627 if (ustconsumer32_data
.channel_monitor_pipe
< 0) {
5629 goto exit_init_data
;
5632 ust64_channel_monitor_pipe
= lttng_pipe_open(0);
5633 if (!ust64_channel_monitor_pipe
) {
5634 ERR("Failed to create 64-bit user space consumer channel monitor pipe");
5636 goto exit_init_data
;
5638 ustconsumer64_data
.channel_monitor_pipe
= lttng_pipe_release_writefd(
5639 ust64_channel_monitor_pipe
);
5640 if (ustconsumer64_data
.channel_monitor_pipe
< 0) {
5642 goto exit_init_data
;
5646 * See if daemon already exist.
5648 if (check_existing_daemon()) {
5649 ERR("Already running daemon.\n");
5651 * We do not goto exit because we must not cleanup()
5652 * because a daemon is already running.
5655 goto exit_init_data
;
5659 * Init UST app hash table. Alloc hash table before this point since
5660 * cleanup() can get called after that point.
5662 if (ust_app_ht_alloc()) {
5663 ERR("Failed to allocate UST app hash table");
5665 goto exit_init_data
;
5669 * Initialize agent app hash table. We allocate the hash table here
5670 * since cleanup() can get called after this point.
5672 if (agent_app_ht_alloc()) {
5673 ERR("Failed to allocate Agent app hash table");
5675 goto exit_init_data
;
5679 * These actions must be executed as root. We do that *after* setting up
5680 * the sockets path because we MUST make the check for another daemon using
5681 * those paths *before* trying to set the kernel consumer sockets and init
5685 if (set_consumer_sockets(&kconsumer_data
)) {
5687 goto exit_init_data
;
5690 /* Setup kernel tracer */
5691 if (!config
.no_kernel
) {
5692 init_kernel_tracer();
5693 if (kernel_tracer_fd
>= 0) {
5694 ret
= syscall_init_table();
5696 ERR("Unable to populate syscall table. "
5697 "Syscall tracing won't work "
5698 "for this session daemon.");
5703 /* Set ulimit for open files */
5706 /* init lttng_fd tracking must be done after set_ulimit. */
5709 if (set_consumer_sockets(&ustconsumer64_data
)) {
5711 goto exit_init_data
;
5714 if (set_consumer_sockets(&ustconsumer32_data
)) {
5716 goto exit_init_data
;
5719 /* Setup the needed unix socket */
5720 if (init_daemon_socket()) {
5722 goto exit_init_data
;
5725 /* Set credentials to socket */
5726 if (is_root
&& set_permissions(config
.rundir
.value
)) {
5728 goto exit_init_data
;
5731 /* Get parent pid if -S, --sig-parent is specified. */
5732 if (config
.sig_parent
) {
5736 /* Setup the kernel pipe for waking up the kernel thread */
5737 if (is_root
&& !config
.no_kernel
) {
5738 if (utils_create_pipe_cloexec(kernel_poll_pipe
)) {
5740 goto exit_init_data
;
5744 /* Setup the thread apps communication pipe. */
5745 if (utils_create_pipe_cloexec(apps_cmd_pipe
)) {
5747 goto exit_init_data
;
5750 /* Setup the thread apps notify communication pipe. */
5751 if (utils_create_pipe_cloexec(apps_cmd_notify_pipe
)) {
5753 goto exit_init_data
;
5756 /* Initialize global buffer per UID and PID registry. */
5757 buffer_reg_init_uid_registry();
5758 buffer_reg_init_pid_registry();
5760 /* Init UST command queue. */
5761 cds_wfcq_init(&ust_cmd_queue
.head
, &ust_cmd_queue
.tail
);
5764 * Get session list pointer. This pointer MUST NOT be free'd. This list
5765 * is statically declared in session.c
5767 session_list_ptr
= session_get_list();
5771 /* Check for the application socket timeout env variable. */
5772 env_app_timeout
= getenv(DEFAULT_APP_SOCKET_TIMEOUT_ENV
);
5773 if (env_app_timeout
) {
5774 config
.app_socket_timeout
= atoi(env_app_timeout
);
5776 config
.app_socket_timeout
= DEFAULT_APP_SOCKET_RW_TIMEOUT
;
5779 ret
= write_pidfile();
5781 ERR("Error in write_pidfile");
5783 goto exit_init_data
;
5785 ret
= write_agent_port();
5787 ERR("Error in write_agent_port");
5789 goto exit_init_data
;
5792 /* Initialize communication library */
5794 /* Initialize TCP timeout values */
5795 lttcomm_inet_init();
5797 if (load_session_init_data(&load_info
) < 0) {
5799 goto exit_init_data
;
5801 load_info
->path
= config
.load_session_path
.value
;
5803 /* Create health-check thread. */
5804 ret
= pthread_create(&health_thread
, default_pthread_attr(),
5805 thread_manage_health
, (void *) NULL
);
5808 PERROR("pthread_create health");
5813 /* notification_thread_data acquires the pipes' read side. */
5814 notification_thread_handle
= notification_thread_handle_create(
5815 ust32_channel_monitor_pipe
,
5816 ust64_channel_monitor_pipe
,
5817 kernel_channel_monitor_pipe
);
5818 if (!notification_thread_handle
) {
5820 ERR("Failed to create notification thread shared data");
5822 goto exit_notification
;
5825 /* Create notification thread. */
5826 ret
= pthread_create(¬ification_thread
, default_pthread_attr(),
5827 thread_notification
, notification_thread_handle
);
5830 PERROR("pthread_create notification");
5833 goto exit_notification
;
5835 notification_thread_running
= true;
5837 /* Create thread to manage the client socket */
5838 ret
= pthread_create(&client_thread
, default_pthread_attr(),
5839 thread_manage_clients
, (void *) NULL
);
5842 PERROR("pthread_create clients");
5848 /* Create thread to dispatch registration */
5849 ret
= pthread_create(&dispatch_thread
, default_pthread_attr(),
5850 thread_dispatch_ust_registration
, (void *) NULL
);
5853 PERROR("pthread_create dispatch");
5859 /* Create thread to manage application registration. */
5860 ret
= pthread_create(®_apps_thread
, default_pthread_attr(),
5861 thread_registration_apps
, (void *) NULL
);
5864 PERROR("pthread_create registration");
5870 /* Create thread to manage application socket */
5871 ret
= pthread_create(&apps_thread
, default_pthread_attr(),
5872 thread_manage_apps
, (void *) NULL
);
5875 PERROR("pthread_create apps");
5881 /* Create thread to manage application notify socket */
5882 ret
= pthread_create(&apps_notify_thread
, default_pthread_attr(),
5883 ust_thread_manage_notify
, (void *) NULL
);
5886 PERROR("pthread_create notify");
5889 goto exit_apps_notify
;
5892 /* Create agent registration thread. */
5893 ret
= pthread_create(&agent_reg_thread
, default_pthread_attr(),
5894 agent_thread_manage_registration
, (void *) NULL
);
5897 PERROR("pthread_create agent");
5900 goto exit_agent_reg
;
5903 /* Don't start this thread if kernel tracing is not requested nor root */
5904 if (is_root
&& !config
.no_kernel
) {
5905 /* Create kernel thread to manage kernel event */
5906 ret
= pthread_create(&kernel_thread
, default_pthread_attr(),
5907 thread_manage_kernel
, (void *) NULL
);
5910 PERROR("pthread_create kernel");
5917 /* Create session loading thread. */
5918 ret
= pthread_create(&load_session_thread
, default_pthread_attr(),
5919 thread_load_session
, load_info
);
5922 PERROR("pthread_create load_session_thread");
5925 goto exit_load_session
;
5929 * This is where we start awaiting program completion (e.g. through
5930 * signal that asks threads to teardown).
5933 ret
= pthread_join(load_session_thread
, &status
);
5936 PERROR("pthread_join load_session_thread");
5941 if (is_root
&& !config
.no_kernel
) {
5942 ret
= pthread_join(kernel_thread
, &status
);
5945 PERROR("pthread_join");
5951 ret
= pthread_join(agent_reg_thread
, &status
);
5954 PERROR("pthread_join agent");
5959 ret
= pthread_join(apps_notify_thread
, &status
);
5962 PERROR("pthread_join apps notify");
5967 ret
= pthread_join(apps_thread
, &status
);
5970 PERROR("pthread_join apps");
5975 ret
= pthread_join(reg_apps_thread
, &status
);
5978 PERROR("pthread_join");
5984 * Join dispatch thread after joining reg_apps_thread to ensure
5985 * we don't leak applications in the queue.
5987 ret
= pthread_join(dispatch_thread
, &status
);
5990 PERROR("pthread_join");
5995 ret
= pthread_join(client_thread
, &status
);
5998 PERROR("pthread_join");
6004 ret
= pthread_join(health_thread
, &status
);
6007 PERROR("pthread_join health thread");
6014 * Wait for all pending call_rcu work to complete before tearing
6015 * down data structures. call_rcu worker may be trying to
6016 * perform lookups in those structures.
6020 * sessiond_cleanup() is called when no other thread is running, except
6021 * the ht_cleanup thread, which is needed to destroy the hash tables.
6023 rcu_thread_online();
6027 * Ensure all prior call_rcu are done. call_rcu callbacks may push
6028 * hash tables to the ht_cleanup thread. Therefore, we ensure that
6029 * the queue is empty before shutting down the clean-up thread.
6034 * The teardown of the notification system is performed after the
6035 * session daemon's teardown in order to allow it to be notified
6036 * of the active session and channels at the moment of the teardown.
6038 if (notification_thread_handle
) {
6039 if (notification_thread_running
) {
6040 notification_thread_command_quit(
6041 notification_thread_handle
);
6042 ret
= pthread_join(notification_thread
, &status
);
6045 PERROR("pthread_join notification thread");
6049 notification_thread_handle_destroy(notification_thread_handle
);
6052 rcu_thread_offline();
6053 rcu_unregister_thread();
6055 ret
= fini_ht_cleanup_thread(&ht_cleanup_thread
);
6059 lttng_pipe_destroy(ust32_channel_monitor_pipe
);
6060 lttng_pipe_destroy(ust64_channel_monitor_pipe
);
6061 lttng_pipe_destroy(kernel_channel_monitor_pipe
);
6064 health_app_destroy(health_sessiond
);
6065 exit_health_sessiond_cleanup
:
6066 exit_create_run_as_worker_cleanup
:
6069 sessiond_cleanup_options();
6071 exit_set_signal_handler
: